initial commit

.gitignore

@@ -0,0 +1,7 @@
+flagged/
+*.pt
+*.png
+*.jpg
+*.mp4
+*.mkv
+gradio_cached_examples/

app.py

@@ -0,0 +1,112 @@
+import gradio as gr
+import cv2
+import requests
+import os
+import torch
+from models.yolo import Model
+
+
+file_urls = [
+    'https://www.dropbox.com/s/b5g97xo901zb3ds/pothole_example.jpg?dl=1',
+    'https://www.dropbox.com/s/86uxlxxlm1iaexa/pothole_screenshot.png?dl=1',
+    'https://www.dropbox.com/s/7sjfwncffg8xej2/video_7.mp4?dl=1'
+]
+
+
+def download_file(url, save_name):
+    url = url
+    if not os.path.exists(save_name):
+        file = requests.get(url)
+        open(save_name, 'wb').write(file.content)
+
+
+for i, url in enumerate(file_urls):
+    if 'mp4' in file_urls[i]:
+        download_file(
+            file_urls[i],
+            f"video.mp4"
+        )
+    else:
+        download_file(
+            file_urls[i],
+            f"image_{i}.jpg"
+        )
+
+model = Model('models/yolov5s.yaml')  
+model.load_state_dict(torch.load('ppe-yolov5.pt')) 
+model.eval()
+path = [['image_0.jpg'], ['image_1.jpg']]
+video_path = [['video.mp4']]
+
+
+def show_preds_image(image_path):
+    image = cv2.imread(image_path)
+    results = model(image)  
+    for i, det in enumerate(results.boxes.xyxy):
+        cv2.rectangle(
+            image,
+            (int(det[0]), int(det[1])),
+            (int(det[2]), int(det[3])),
+            color=(0, 0, 255),
+            thickness=2,
+            lineType=cv2.LINE_AA
+        )
+    return cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
+
+
+inputs_image = [
+    gr.components.Image(type="filepath", label="Input Image"),
+]
+outputs_image = [
+    gr.components.Image(type="numpy", label="Output Image"),
+]
+interface_image = gr.Interface(
+    fn=show_preds_image,
+    inputs=inputs_image,
+    outputs=outputs_image,
+    title="PPE Detector",
+    examples=path,
+    cache_examples=False,
+)
+
+
+def show_preds_video(video_path):
+    cap = cv2.VideoCapture(video_path)
+    while (cap.isOpened()):
+        ret, frame = cap.read()
+        if ret:
+            frame_copy = frame.copy()
+            outputs = model.predict(source=frame)
+            results = outputs[0].cpu().numpy()
+            for i, det in enumerate(results.boxes.xyxy):
+                cv2.rectangle(
+                    frame_copy,
+                    (int(det[0]), int(det[1])),
+                    (int(det[2]), int(det[3])),
+                    color=(0, 0, 255),
+                    thickness=2,
+                    lineType=cv2.LINE_AA
+                )
+            yield cv2.cvtColor(frame_copy, cv2.COLOR_BGR2RGB)
+
+
+inputs_video = [
+    gr.components.Video(type="filepath", label="Input Video"),
+
+]
+outputs_video = [
+    gr.components.Image(type="numpy", label="Output Image"),
+]
+interface_video = gr.Interface(
+    fn=show_preds_video,
+    inputs=inputs_video,
+    outputs=outputs_video,
+    title="PPE Detector",
+    examples=video_path,
+    cache_examples=False,
+)
+
+gr.TabbedInterface(
+    [interface_image, interface_video],
+    tab_names=['Image inference', 'Video inference']
+).queue().launch()

detect.py

@@ -0,0 +1,285 @@
+# YOLOv5 🚀 by Ultralytics, AGPL-3.0 license
+"""
+Run YOLOv5 detection inference on images, videos, directories, globs, YouTube, webcam, streams, etc.
+
+Usage - sources:
+    $ python detect.py --weights yolov5s.pt --source 0                               # webcam
+                                                     img.jpg                         # image
+                                                     vid.mp4                         # video
+                                                     screen                          # screenshot
+                                                     path/                           # directory
+                                                     list.txt                        # list of images
+                                                     list.streams                    # list of streams
+                                                     'path/*.jpg'                    # glob
+                                                     'https://youtu.be/LNwODJXcvt4'  # YouTube
+                                                     'rtsp://example.com/media.mp4'  # RTSP, RTMP, HTTP stream
+
+Usage - formats:
+    $ python detect.py --weights yolov5s.pt                 # PyTorch
+                                 yolov5s.torchscript        # TorchScript
+                                 yolov5s.onnx               # ONNX Runtime or OpenCV DNN with --dnn
+                                 yolov5s_openvino_model     # OpenVINO
+                                 yolov5s.engine             # TensorRT
+                                 yolov5s.mlmodel            # CoreML (macOS-only)
+                                 yolov5s_saved_model        # TensorFlow SavedModel
+                                 yolov5s.pb                 # TensorFlow GraphDef
+                                 yolov5s.tflite             # TensorFlow Lite
+                                 yolov5s_edgetpu.tflite     # TensorFlow Edge TPU
+                                 yolov5s_paddle_model       # PaddlePaddle
+"""
+
+import argparse
+import csv
+import os
+import platform
+import sys
+from pathlib import Path
+
+import torch
+
+FILE = Path(__file__).resolve()
+ROOT = FILE.parents[0]  # YOLOv5 root directory
+if str(ROOT) not in sys.path:
+    sys.path.append(str(ROOT))  # add ROOT to PATH
+ROOT = Path(os.path.relpath(ROOT, Path.cwd()))  # relative
+
+from ultralytics.utils.plotting import Annotator, colors, save_one_box
+
+from models.common import DetectMultiBackend
+from utils.dataloaders import IMG_FORMATS, VID_FORMATS, LoadImages, LoadScreenshots, LoadStreams
+from utils.general import (LOGGER, Profile, check_file, check_img_size, check_imshow, check_requirements, colorstr, cv2,
+                           increment_path, non_max_suppression, print_args, scale_boxes, strip_optimizer, xyxy2xywh)
+from utils.torch_utils import select_device, smart_inference_mode
+
+
+@smart_inference_mode()
+def run(
+        weights=ROOT / 'yolov5s.pt',  # model path or triton URL
+        source=ROOT / 'data/images',  # file/dir/URL/glob/screen/0(webcam)
+        data=ROOT / 'data/coco128.yaml',  # dataset.yaml path
+        imgsz=(640, 640),  # inference size (height, width)
+        conf_thres=0.25,  # confidence threshold
+        iou_thres=0.45,  # NMS IOU threshold
+        max_det=1000,  # maximum detections per image
+        device='',  # cuda device, i.e. 0 or 0,1,2,3 or cpu
+        view_img=False,  # show results
+        save_txt=False,  # save results to *.txt
+        save_csv=False,  # save results in CSV format
+        save_conf=False,  # save confidences in --save-txt labels
+        save_crop=False,  # save cropped prediction boxes
+        nosave=False,  # do not save images/videos
+        classes=None,  # filter by class: --class 0, or --class 0 2 3
+        agnostic_nms=False,  # class-agnostic NMS
+        augment=False,  # augmented inference
+        visualize=False,  # visualize features
+        update=False,  # update all models
+        project=ROOT / 'runs/detect',  # save results to project/name
+        name='exp',  # save results to project/name
+        exist_ok=False,  # existing project/name ok, do not increment
+        line_thickness=3,  # bounding box thickness (pixels)
+        hide_labels=False,  # hide labels
+        hide_conf=False,  # hide confidences
+        half=False,  # use FP16 half-precision inference
+        dnn=False,  # use OpenCV DNN for ONNX inference
+        vid_stride=1,  # video frame-rate stride
+):
+    source = str(source)
+    save_img = not nosave and not source.endswith('.txt')  # save inference images
+    is_file = Path(source).suffix[1:] in (IMG_FORMATS + VID_FORMATS)
+    is_url = source.lower().startswith(('rtsp://', 'rtmp://', 'http://', 'https://'))
+    webcam = source.isnumeric() or source.endswith('.streams') or (is_url and not is_file)
+    screenshot = source.lower().startswith('screen')
+    if is_url and is_file:
+        source = check_file(source)  # download
+
+    # Directories
+    save_dir = increment_path(Path(project) / name, exist_ok=exist_ok)  # increment run
+    (save_dir / 'labels' if save_txt else save_dir).mkdir(parents=True, exist_ok=True)  # make dir
+
+    # Load model
+    device = select_device(device)
+    model = DetectMultiBackend(weights, device=device, dnn=dnn, data=data, fp16=half)
+    stride, names, pt = model.stride, model.names, model.pt
+    imgsz = check_img_size(imgsz, s=stride)  # check image size
+
+    # Dataloader
+    bs = 1  # batch_size
+    if webcam:
+        view_img = check_imshow(warn=True)
+        dataset = LoadStreams(source, img_size=imgsz, stride=stride, auto=pt, vid_stride=vid_stride)
+        bs = len(dataset)
+    elif screenshot:
+        dataset = LoadScreenshots(source, img_size=imgsz, stride=stride, auto=pt)
+    else:
+        dataset = LoadImages(source, img_size=imgsz, stride=stride, auto=pt, vid_stride=vid_stride)
+    vid_path, vid_writer = [None] * bs, [None] * bs
+
+    # Run inference
+    model.warmup(imgsz=(1 if pt or model.triton else bs, 3, *imgsz))  # warmup
+    seen, windows, dt = 0, [], (Profile(), Profile(), Profile())
+    for path, im, im0s, vid_cap, s in dataset:
+        with dt[0]:
+            im = torch.from_numpy(im).to(model.device)
+            im = im.half() if model.fp16 else im.float()  # uint8 to fp16/32
+            im /= 255  # 0 - 255 to 0.0 - 1.0
+            if len(im.shape) == 3:
+                im = im[None]  # expand for batch dim
+
+        # Inference
+        with dt[1]:
+            visualize = increment_path(save_dir / Path(path).stem, mkdir=True) if visualize else False
+            pred = model(im, augment=augment, visualize=visualize)
+
+        # NMS
+        with dt[2]:
+            pred = non_max_suppression(pred, conf_thres, iou_thres, classes, agnostic_nms, max_det=max_det)
+
+        # Second-stage classifier (optional)
+        # pred = utils.general.apply_classifier(pred, classifier_model, im, im0s)
+
+        # Define the path for the CSV file
+        csv_path = save_dir / 'predictions.csv'
+
+        # Create or append to the CSV file
+        def write_to_csv(image_name, prediction, confidence):
+            data = {'Image Name': image_name, 'Prediction': prediction, 'Confidence': confidence}
+            with open(csv_path, mode='a', newline='') as f:
+                writer = csv.DictWriter(f, fieldnames=data.keys())
+                if not csv_path.is_file():
+                    writer.writeheader()
+                writer.writerow(data)
+
+        # Process predictions
+        for i, det in enumerate(pred):  # per image
+            seen += 1
+            if webcam:  # batch_size >= 1
+                p, im0, frame = path[i], im0s[i].copy(), dataset.count
+                s += f'{i}: '
+            else:
+                p, im0, frame = path, im0s.copy(), getattr(dataset, 'frame', 0)
+
+            p = Path(p)  # to Path
+            save_path = str(save_dir / p.name)  # im.jpg
+            txt_path = str(save_dir / 'labels' / p.stem) + ('' if dataset.mode == 'image' else f'_{frame}')  # im.txt
+            s += '%gx%g ' % im.shape[2:]  # print string
+            gn = torch.tensor(im0.shape)[[1, 0, 1, 0]]  # normalization gain whwh
+            imc = im0.copy() if save_crop else im0  # for save_crop
+            annotator = Annotator(im0, line_width=line_thickness, example=str(names))
+            if len(det):
+                # Rescale boxes from img_size to im0 size
+                det[:, :4] = scale_boxes(im.shape[2:], det[:, :4], im0.shape).round()
+
+                # Print results
+                for c in det[:, 5].unique():
+                    n = (det[:, 5] == c).sum()  # detections per class
+                    s += f"{n} {names[int(c)]}{'s' * (n > 1)}, "  # add to string
+
+                # Write results
+                for *xyxy, conf, cls in reversed(det):
+                    c = int(cls)  # integer class
+                    label = names[c] if hide_conf else f'{names[c]}'
+                    confidence = float(conf)
+                    confidence_str = f'{confidence:.2f}'
+
+                    if save_csv:
+                        write_to_csv(p.name, label, confidence_str)
+
+                    if save_txt:  # Write to file
+                        xywh = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) / gn).view(-1).tolist()  # normalized xywh
+                        line = (cls, *xywh, conf) if save_conf else (cls, *xywh)  # label format
+                        with open(f'{txt_path}.txt', 'a') as f:
+                            f.write(('%g ' * len(line)).rstrip() % line + '\n')
+
+                    if save_img or save_crop or view_img:  # Add bbox to image
+                        c = int(cls)  # integer class
+                        label = None if hide_labels else (names[c] if hide_conf else f'{names[c]} {conf:.2f}')
+                        annotator.box_label(xyxy, label, color=colors(c, True))
+                    if save_crop:
+                        save_one_box(xyxy, imc, file=save_dir / 'crops' / names[c] / f'{p.stem}.jpg', BGR=True)
+
+            # Stream results
+            im0 = annotator.result()
+            if view_img:
+                if platform.system() == 'Linux' and p not in windows:
+                    windows.append(p)
+                    cv2.namedWindow(str(p), cv2.WINDOW_NORMAL | cv2.WINDOW_KEEPRATIO)  # allow window resize (Linux)
+                    cv2.resizeWindow(str(p), im0.shape[1], im0.shape[0])
+                cv2.imshow(str(p), im0)
+                cv2.waitKey(1)  # 1 millisecond
+
+            # Save results (image with detections)
+            if save_img:
+                if dataset.mode == 'image':
+                    cv2.imwrite(save_path, im0)
+                else:  # 'video' or 'stream'
+                    if vid_path[i] != save_path:  # new video
+                        vid_path[i] = save_path
+                        if isinstance(vid_writer[i], cv2.VideoWriter):
+                            vid_writer[i].release()  # release previous video writer
+                        if vid_cap:  # video
+                            fps = vid_cap.get(cv2.CAP_PROP_FPS)
+                            w = int(vid_cap.get(cv2.CAP_PROP_FRAME_WIDTH))
+                            h = int(vid_cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
+                        else:  # stream
+                            fps, w, h = 30, im0.shape[1], im0.shape[0]
+                        save_path = str(Path(save_path).with_suffix('.mp4'))  # force *.mp4 suffix on results videos
+                        vid_writer[i] = cv2.VideoWriter(save_path, cv2.VideoWriter_fourcc(*'mp4v'), fps, (w, h))
+                    vid_writer[i].write(im0)
+
+        # Print time (inference-only)
+        LOGGER.info(f"{s}{'' if len(det) else '(no detections), '}{dt[1].dt * 1E3:.1f}ms")
+
+    # Print results
+    t = tuple(x.t / seen * 1E3 for x in dt)  # speeds per image
+    LOGGER.info(f'Speed: %.1fms pre-process, %.1fms inference, %.1fms NMS per image at shape {(1, 3, *imgsz)}' % t)
+    if save_txt or save_img:
+        s = f"\n{len(list(save_dir.glob('labels/*.txt')))} labels saved to {save_dir / 'labels'}" if save_txt else ''
+        LOGGER.info(f"Results saved to {colorstr('bold', save_dir)}{s}")
+    if update:
+        strip_optimizer(weights[0])  # update model (to fix SourceChangeWarning)
+
+
+def parse_opt():
+    parser = argparse.ArgumentParser()
+    parser.add_argument('--weights', nargs='+', type=str, default=ROOT / 'yolov5s.pt', help='model path or triton URL')
+    parser.add_argument('--source', type=str, default=ROOT / 'data/images', help='file/dir/URL/glob/screen/0(webcam)')
+    parser.add_argument('--data', type=str, default=ROOT / 'data/coco128.yaml', help='(optional) dataset.yaml path')
+    parser.add_argument('--imgsz', '--img', '--img-size', nargs='+', type=int, default=[640], help='inference size h,w')
+    parser.add_argument('--conf-thres', type=float, default=0.25, help='confidence threshold')
+    parser.add_argument('--iou-thres', type=float, default=0.45, help='NMS IoU threshold')
+    parser.add_argument('--max-det', type=int, default=1000, help='maximum detections per image')
+    parser.add_argument('--device', default='', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')
+    parser.add_argument('--view-img', action='store_true', help='show results')
+    parser.add_argument('--save-txt', action='store_true', help='save results to *.txt')
+    parser.add_argument('--save-csv', action='store_true', help='save results in CSV format')
+    parser.add_argument('--save-conf', action='store_true', help='save confidences in --save-txt labels')
+    parser.add_argument('--save-crop', action='store_true', help='save cropped prediction boxes')
+    parser.add_argument('--nosave', action='store_true', help='do not save images/videos')
+    parser.add_argument('--classes', nargs='+', type=int, help='filter by class: --classes 0, or --classes 0 2 3')
+    parser.add_argument('--agnostic-nms', action='store_true', help='class-agnostic NMS')
+    parser.add_argument('--augment', action='store_true', help='augmented inference')
+    parser.add_argument('--visualize', action='store_true', help='visualize features')
+    parser.add_argument('--update', action='store_true', help='update all models')
+    parser.add_argument('--project', default=ROOT / 'runs/detect', help='save results to project/name')
+    parser.add_argument('--name', default='exp', help='save results to project/name')
+    parser.add_argument('--exist-ok', action='store_true', help='existing project/name ok, do not increment')
+    parser.add_argument('--line-thickness', default=3, type=int, help='bounding box thickness (pixels)')
+    parser.add_argument('--hide-labels', default=False, action='store_true', help='hide labels')
+    parser.add_argument('--hide-conf', default=False, action='store_true', help='hide confidences')
+    parser.add_argument('--half', action='store_true', help='use FP16 half-precision inference')
+    parser.add_argument('--dnn', action='store_true', help='use OpenCV DNN for ONNX inference')
+    parser.add_argument('--vid-stride', type=int, default=1, help='video frame-rate stride')
+    opt = parser.parse_args()
+    opt.imgsz *= 2 if len(opt.imgsz) == 1 else 1  # expand
+    print_args(vars(opt))
+    return opt
+
+
+def main(opt):
+    check_requirements(ROOT / 'requirements.txt', exclude=('tensorboard', 'thop'))
+    run(**vars(opt))
+
+
+if __name__ == '__main__':
+    opt = parse_opt()
+    main(opt)

models/common.py

@@ -0,0 +1,883 @@
+# YOLOv5 🚀 by Ultralytics, AGPL-3.0 license
+"""
+Common modules
+"""
+
+import ast
+import contextlib
+import json
+import math
+import platform
+import warnings
+import zipfile
+from collections import OrderedDict, namedtuple
+from copy import copy
+from pathlib import Path
+from urllib.parse import urlparse
+
+import cv2
+import numpy as np
+import pandas as pd
+import requests
+import torch
+import torch.nn as nn
+from PIL import Image
+from torch.cuda import amp
+
+# Import 'ultralytics' package or install if if missing
+try:
+    import ultralytics
+
+    assert hasattr(ultralytics, '__version__')  # verify package is not directory
+except (ImportError, AssertionError):
+    import os
+
+    os.system('pip install -U ultralytics')
+    import ultralytics
+
+from ultralytics.utils.plotting import Annotator, colors, save_one_box
+
+from utils import TryExcept
+from utils.dataloaders import exif_transpose, letterbox
+from utils.general import (LOGGER, ROOT, Profile, check_requirements, check_suffix, check_version, colorstr,
+                           increment_path, is_jupyter, make_divisible, non_max_suppression, scale_boxes, xywh2xyxy,
+                           xyxy2xywh, yaml_load)
+from utils.torch_utils import copy_attr, smart_inference_mode
+
+
+def autopad(k, p=None, d=1):  # kernel, padding, dilation
+    # Pad to 'same' shape outputs
+    if d > 1:
+        k = d * (k - 1) + 1 if isinstance(k, int) else [d * (x - 1) + 1 for x in k]  # actual kernel-size
+    if p is None:
+        p = k // 2 if isinstance(k, int) else [x // 2 for x in k]  # auto-pad
+    return p
+
+
+class Conv(nn.Module):
+    # Standard convolution with args(ch_in, ch_out, kernel, stride, padding, groups, dilation, activation)
+    default_act = nn.SiLU()  # default activation
+
+    def __init__(self, c1, c2, k=1, s=1, p=None, g=1, d=1, act=True):
+        super().__init__()
+        self.conv = nn.Conv2d(c1, c2, k, s, autopad(k, p, d), groups=g, dilation=d, bias=False)
+        self.bn = nn.BatchNorm2d(c2)
+        self.act = self.default_act if act is True else act if isinstance(act, nn.Module) else nn.Identity()
+
+    def forward(self, x):
+        return self.act(self.bn(self.conv(x)))
+
+    def forward_fuse(self, x):
+        return self.act(self.conv(x))
+
+
+class DWConv(Conv):
+    # Depth-wise convolution
+    def __init__(self, c1, c2, k=1, s=1, d=1, act=True):  # ch_in, ch_out, kernel, stride, dilation, activation
+        super().__init__(c1, c2, k, s, g=math.gcd(c1, c2), d=d, act=act)
+
+
+class DWConvTranspose2d(nn.ConvTranspose2d):
+    # Depth-wise transpose convolution
+    def __init__(self, c1, c2, k=1, s=1, p1=0, p2=0):  # ch_in, ch_out, kernel, stride, padding, padding_out
+        super().__init__(c1, c2, k, s, p1, p2, groups=math.gcd(c1, c2))
+
+
+class TransformerLayer(nn.Module):
+    # Transformer layer https://arxiv.org/abs/2010.11929 (LayerNorm layers removed for better performance)
+    def __init__(self, c, num_heads):
+        super().__init__()
+        self.q = nn.Linear(c, c, bias=False)
+        self.k = nn.Linear(c, c, bias=False)
+        self.v = nn.Linear(c, c, bias=False)
+        self.ma = nn.MultiheadAttention(embed_dim=c, num_heads=num_heads)
+        self.fc1 = nn.Linear(c, c, bias=False)
+        self.fc2 = nn.Linear(c, c, bias=False)
+
+    def forward(self, x):
+        x = self.ma(self.q(x), self.k(x), self.v(x))[0] + x
+        x = self.fc2(self.fc1(x)) + x
+        return x
+
+
+class TransformerBlock(nn.Module):
+    # Vision Transformer https://arxiv.org/abs/2010.11929
+    def __init__(self, c1, c2, num_heads, num_layers):
+        super().__init__()
+        self.conv = None
+        if c1 != c2:
+            self.conv = Conv(c1, c2)
+        self.linear = nn.Linear(c2, c2)  # learnable position embedding
+        self.tr = nn.Sequential(*(TransformerLayer(c2, num_heads) for _ in range(num_layers)))
+        self.c2 = c2
+
+    def forward(self, x):
+        if self.conv is not None:
+            x = self.conv(x)
+        b, _, w, h = x.shape
+        p = x.flatten(2).permute(2, 0, 1)
+        return self.tr(p + self.linear(p)).permute(1, 2, 0).reshape(b, self.c2, w, h)
+
+
+class Bottleneck(nn.Module):
+    # Standard bottleneck
+    def __init__(self, c1, c2, shortcut=True, g=1, e=0.5):  # ch_in, ch_out, shortcut, groups, expansion
+        super().__init__()
+        c_ = int(c2 * e)  # hidden channels
+        self.cv1 = Conv(c1, c_, 1, 1)
+        self.cv2 = Conv(c_, c2, 3, 1, g=g)
+        self.add = shortcut and c1 == c2
+
+    def forward(self, x):
+        return x + self.cv2(self.cv1(x)) if self.add else self.cv2(self.cv1(x))
+
+
+class BottleneckCSP(nn.Module):
+    # CSP Bottleneck https://github.com/WongKinYiu/CrossStagePartialNetworks
+    def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5):  # ch_in, ch_out, number, shortcut, groups, expansion
+        super().__init__()
+        c_ = int(c2 * e)  # hidden channels
+        self.cv1 = Conv(c1, c_, 1, 1)
+        self.cv2 = nn.Conv2d(c1, c_, 1, 1, bias=False)
+        self.cv3 = nn.Conv2d(c_, c_, 1, 1, bias=False)
+        self.cv4 = Conv(2 * c_, c2, 1, 1)
+        self.bn = nn.BatchNorm2d(2 * c_)  # applied to cat(cv2, cv3)
+        self.act = nn.SiLU()
+        self.m = nn.Sequential(*(Bottleneck(c_, c_, shortcut, g, e=1.0) for _ in range(n)))
+
+    def forward(self, x):
+        y1 = self.cv3(self.m(self.cv1(x)))
+        y2 = self.cv2(x)
+        return self.cv4(self.act(self.bn(torch.cat((y1, y2), 1))))
+
+
+class CrossConv(nn.Module):
+    # Cross Convolution Downsample
+    def __init__(self, c1, c2, k=3, s=1, g=1, e=1.0, shortcut=False):
+        # ch_in, ch_out, kernel, stride, groups, expansion, shortcut
+        super().__init__()
+        c_ = int(c2 * e)  # hidden channels
+        self.cv1 = Conv(c1, c_, (1, k), (1, s))
+        self.cv2 = Conv(c_, c2, (k, 1), (s, 1), g=g)
+        self.add = shortcut and c1 == c2
+
+    def forward(self, x):
+        return x + self.cv2(self.cv1(x)) if self.add else self.cv2(self.cv1(x))
+
+
+class C3(nn.Module):
+    # CSP Bottleneck with 3 convolutions
+    def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5):  # ch_in, ch_out, number, shortcut, groups, expansion
+        super().__init__()
+        c_ = int(c2 * e)  # hidden channels
+        self.cv1 = Conv(c1, c_, 1, 1)
+        self.cv2 = Conv(c1, c_, 1, 1)
+        self.cv3 = Conv(2 * c_, c2, 1)  # optional act=FReLU(c2)
+        self.m = nn.Sequential(*(Bottleneck(c_, c_, shortcut, g, e=1.0) for _ in range(n)))
+
+    def forward(self, x):
+        return self.cv3(torch.cat((self.m(self.cv1(x)), self.cv2(x)), 1))
+
+
+class C3x(C3):
+    # C3 module with cross-convolutions
+    def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5):
+        super().__init__(c1, c2, n, shortcut, g, e)
+        c_ = int(c2 * e)
+        self.m = nn.Sequential(*(CrossConv(c_, c_, 3, 1, g, 1.0, shortcut) for _ in range(n)))
+
+
+class C3TR(C3):
+    # C3 module with TransformerBlock()
+    def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5):
+        super().__init__(c1, c2, n, shortcut, g, e)
+        c_ = int(c2 * e)
+        self.m = TransformerBlock(c_, c_, 4, n)
+
+
+class C3SPP(C3):
+    # C3 module with SPP()
+    def __init__(self, c1, c2, k=(5, 9, 13), n=1, shortcut=True, g=1, e=0.5):
+        super().__init__(c1, c2, n, shortcut, g, e)
+        c_ = int(c2 * e)
+        self.m = SPP(c_, c_, k)
+
+
+class C3Ghost(C3):
+    # C3 module with GhostBottleneck()
+    def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5):
+        super().__init__(c1, c2, n, shortcut, g, e)
+        c_ = int(c2 * e)  # hidden channels
+        self.m = nn.Sequential(*(GhostBottleneck(c_, c_) for _ in range(n)))
+
+
+class SPP(nn.Module):
+    # Spatial Pyramid Pooling (SPP) layer https://arxiv.org/abs/1406.4729
+    def __init__(self, c1, c2, k=(5, 9, 13)):
+        super().__init__()
+        c_ = c1 // 2  # hidden channels
+        self.cv1 = Conv(c1, c_, 1, 1)
+        self.cv2 = Conv(c_ * (len(k) + 1), c2, 1, 1)
+        self.m = nn.ModuleList([nn.MaxPool2d(kernel_size=x, stride=1, padding=x // 2) for x in k])
+
+    def forward(self, x):
+        x = self.cv1(x)
+        with warnings.catch_warnings():
+            warnings.simplefilter('ignore')  # suppress torch 1.9.0 max_pool2d() warning
+            return self.cv2(torch.cat([x] + [m(x) for m in self.m], 1))
+
+
+class SPPF(nn.Module):
+    # Spatial Pyramid Pooling - Fast (SPPF) layer for YOLOv5 by Glenn Jocher
+    def __init__(self, c1, c2, k=5):  # equivalent to SPP(k=(5, 9, 13))
+        super().__init__()
+        c_ = c1 // 2  # hidden channels
+        self.cv1 = Conv(c1, c_, 1, 1)
+        self.cv2 = Conv(c_ * 4, c2, 1, 1)
+        self.m = nn.MaxPool2d(kernel_size=k, stride=1, padding=k // 2)
+
+    def forward(self, x):
+        x = self.cv1(x)
+        with warnings.catch_warnings():
+            warnings.simplefilter('ignore')  # suppress torch 1.9.0 max_pool2d() warning
+            y1 = self.m(x)
+            y2 = self.m(y1)
+            return self.cv2(torch.cat((x, y1, y2, self.m(y2)), 1))
+
+
+class Focus(nn.Module):
+    # Focus wh information into c-space
+    def __init__(self, c1, c2, k=1, s=1, p=None, g=1, act=True):  # ch_in, ch_out, kernel, stride, padding, groups
+        super().__init__()
+        self.conv = Conv(c1 * 4, c2, k, s, p, g, act=act)
+        # self.contract = Contract(gain=2)
+
+    def forward(self, x):  # x(b,c,w,h) -> y(b,4c,w/2,h/2)
+        return self.conv(torch.cat((x[..., ::2, ::2], x[..., 1::2, ::2], x[..., ::2, 1::2], x[..., 1::2, 1::2]), 1))
+        # return self.conv(self.contract(x))
+
+
+class GhostConv(nn.Module):
+    # Ghost Convolution https://github.com/huawei-noah/ghostnet
+    def __init__(self, c1, c2, k=1, s=1, g=1, act=True):  # ch_in, ch_out, kernel, stride, groups
+        super().__init__()
+        c_ = c2 // 2  # hidden channels
+        self.cv1 = Conv(c1, c_, k, s, None, g, act=act)
+        self.cv2 = Conv(c_, c_, 5, 1, None, c_, act=act)
+
+    def forward(self, x):
+        y = self.cv1(x)
+        return torch.cat((y, self.cv2(y)), 1)
+
+
+class GhostBottleneck(nn.Module):
+    # Ghost Bottleneck https://github.com/huawei-noah/ghostnet
+    def __init__(self, c1, c2, k=3, s=1):  # ch_in, ch_out, kernel, stride
+        super().__init__()
+        c_ = c2 // 2
+        self.conv = nn.Sequential(
+            GhostConv(c1, c_, 1, 1),  # pw
+            DWConv(c_, c_, k, s, act=False) if s == 2 else nn.Identity(),  # dw
+            GhostConv(c_, c2, 1, 1, act=False))  # pw-linear
+        self.shortcut = nn.Sequential(DWConv(c1, c1, k, s, act=False), Conv(c1, c2, 1, 1,
+                                                                            act=False)) if s == 2 else nn.Identity()
+
+    def forward(self, x):
+        return self.conv(x) + self.shortcut(x)
+
+
+class Contract(nn.Module):
+    # Contract width-height into channels, i.e. x(1,64,80,80) to x(1,256,40,40)
+    def __init__(self, gain=2):
+        super().__init__()
+        self.gain = gain
+
+    def forward(self, x):
+        b, c, h, w = x.size()  # assert (h / s == 0) and (W / s == 0), 'Indivisible gain'
+        s = self.gain
+        x = x.view(b, c, h // s, s, w // s, s)  # x(1,64,40,2,40,2)
+        x = x.permute(0, 3, 5, 1, 2, 4).contiguous()  # x(1,2,2,64,40,40)
+        return x.view(b, c * s * s, h // s, w // s)  # x(1,256,40,40)
+
+
+class Expand(nn.Module):
+    # Expand channels into width-height, i.e. x(1,64,80,80) to x(1,16,160,160)
+    def __init__(self, gain=2):
+        super().__init__()
+        self.gain = gain
+
+    def forward(self, x):
+        b, c, h, w = x.size()  # assert C / s ** 2 == 0, 'Indivisible gain'
+        s = self.gain
+        x = x.view(b, s, s, c // s ** 2, h, w)  # x(1,2,2,16,80,80)
+        x = x.permute(0, 3, 4, 1, 5, 2).contiguous()  # x(1,16,80,2,80,2)
+        return x.view(b, c // s ** 2, h * s, w * s)  # x(1,16,160,160)
+
+
+class Concat(nn.Module):
+    # Concatenate a list of tensors along dimension
+    def __init__(self, dimension=1):
+        super().__init__()
+        self.d = dimension
+
+    def forward(self, x):
+        return torch.cat(x, self.d)
+
+
+class DetectMultiBackend(nn.Module):
+    # YOLOv5 MultiBackend class for python inference on various backends
+    def __init__(self, weights='yolov5s.pt', device=torch.device('cpu'), dnn=False, data=None, fp16=False, fuse=True):
+        # Usage:
+        #   PyTorch:              weights = *.pt
+        #   TorchScript:                    *.torchscript
+        #   ONNX Runtime:                   *.onnx
+        #   ONNX OpenCV DNN:                *.onnx --dnn
+        #   OpenVINO:                       *_openvino_model
+        #   CoreML:                         *.mlmodel
+        #   TensorRT:                       *.engine
+        #   TensorFlow SavedModel:          *_saved_model
+        #   TensorFlow GraphDef:            *.pb
+        #   TensorFlow Lite:                *.tflite
+        #   TensorFlow Edge TPU:            *_edgetpu.tflite
+        #   PaddlePaddle:                   *_paddle_model
+        from models.experimental import attempt_download, attempt_load  # scoped to avoid circular import
+
+        super().__init__()
+        w = str(weights[0] if isinstance(weights, list) else weights)
+        pt, jit, onnx, xml, engine, coreml, saved_model, pb, tflite, edgetpu, tfjs, paddle, triton = self._model_type(w)
+        fp16 &= pt or jit or onnx or engine or triton  # FP16
+        nhwc = coreml or saved_model or pb or tflite or edgetpu  # BHWC formats (vs torch BCWH)
+        stride = 32  # default stride
+        cuda = torch.cuda.is_available() and device.type != 'cpu'  # use CUDA
+        if not (pt or triton):
+            w = attempt_download(w)  # download if not local
+
+        if pt:  # PyTorch
+            model = attempt_load(weights if isinstance(weights, list) else w, device=device, inplace=True, fuse=fuse)
+            stride = max(int(model.stride.max()), 32)  # model stride
+            names = model.module.names if hasattr(model, 'module') else model.names  # get class names
+            model.half() if fp16 else model.float()
+            self.model = model  # explicitly assign for to(), cpu(), cuda(), half()
+        elif jit:  # TorchScript
+            LOGGER.info(f'Loading {w} for TorchScript inference...')
+            extra_files = {'config.txt': ''}  # model metadata
+            model = torch.jit.load(w, _extra_files=extra_files, map_location=device)
+            model.half() if fp16 else model.float()
+            if extra_files['config.txt']:  # load metadata dict
+                d = json.loads(extra_files['config.txt'],
+                               object_hook=lambda d: {
+                                   int(k) if k.isdigit() else k: v
+                                   for k, v in d.items()})
+                stride, names = int(d['stride']), d['names']
+        elif dnn:  # ONNX OpenCV DNN
+            LOGGER.info(f'Loading {w} for ONNX OpenCV DNN inference...')
+            check_requirements('opencv-python>=4.5.4')
+            net = cv2.dnn.readNetFromONNX(w)
+        elif onnx:  # ONNX Runtime
+            LOGGER.info(f'Loading {w} for ONNX Runtime inference...')
+            check_requirements(('onnx', 'onnxruntime-gpu' if cuda else 'onnxruntime'))
+            import onnxruntime
+            providers = ['CUDAExecutionProvider', 'CPUExecutionProvider'] if cuda else ['CPUExecutionProvider']
+            session = onnxruntime.InferenceSession(w, providers=providers)
+            output_names = [x.name for x in session.get_outputs()]
+            meta = session.get_modelmeta().custom_metadata_map  # metadata
+            if 'stride' in meta:
+                stride, names = int(meta['stride']), eval(meta['names'])
+        elif xml:  # OpenVINO
+            LOGGER.info(f'Loading {w} for OpenVINO inference...')
+            check_requirements('openvino>=2023.0')  # requires openvino-dev: https://pypi.org/project/openvino-dev/
+            from openvino.runtime import Core, Layout, get_batch
+            core = Core()
+            if not Path(w).is_file():  # if not *.xml
+                w = next(Path(w).glob('*.xml'))  # get *.xml file from *_openvino_model dir
+            ov_model = core.read_model(model=w, weights=Path(w).with_suffix('.bin'))
+            if ov_model.get_parameters()[0].get_layout().empty:
+                ov_model.get_parameters()[0].set_layout(Layout('NCHW'))
+            batch_dim = get_batch(ov_model)
+            if batch_dim.is_static:
+                batch_size = batch_dim.get_length()
+            ov_compiled_model = core.compile_model(ov_model, device_name='AUTO')  # AUTO selects best available device
+            stride, names = self._load_metadata(Path(w).with_suffix('.yaml'))  # load metadata
+        elif engine:  # TensorRT
+            LOGGER.info(f'Loading {w} for TensorRT inference...')
+            import tensorrt as trt  # https://developer.nvidia.com/nvidia-tensorrt-download
+            check_version(trt.__version__, '7.0.0', hard=True)  # require tensorrt>=7.0.0
+            if device.type == 'cpu':
+                device = torch.device('cuda:0')
+            Binding = namedtuple('Binding', ('name', 'dtype', 'shape', 'data', 'ptr'))
+            logger = trt.Logger(trt.Logger.INFO)
+            with open(w, 'rb') as f, trt.Runtime(logger) as runtime:
+                model = runtime.deserialize_cuda_engine(f.read())
+            context = model.create_execution_context()
+            bindings = OrderedDict()
+            output_names = []
+            fp16 = False  # default updated below
+            dynamic = False
+            for i in range(model.num_bindings):
+                name = model.get_binding_name(i)
+                dtype = trt.nptype(model.get_binding_dtype(i))
+                if model.binding_is_input(i):
+                    if -1 in tuple(model.get_binding_shape(i)):  # dynamic
+                        dynamic = True
+                        context.set_binding_shape(i, tuple(model.get_profile_shape(0, i)[2]))
+                    if dtype == np.float16:
+                        fp16 = True
+                else:  # output
+                    output_names.append(name)
+                shape = tuple(context.get_binding_shape(i))
+                im = torch.from_numpy(np.empty(shape, dtype=dtype)).to(device)
+                bindings[name] = Binding(name, dtype, shape, im, int(im.data_ptr()))
+            binding_addrs = OrderedDict((n, d.ptr) for n, d in bindings.items())
+            batch_size = bindings['images'].shape[0]  # if dynamic, this is instead max batch size
+        elif coreml:  # CoreML
+            LOGGER.info(f'Loading {w} for CoreML inference...')
+            import coremltools as ct
+            model = ct.models.MLModel(w)
+        elif saved_model:  # TF SavedModel
+            LOGGER.info(f'Loading {w} for TensorFlow SavedModel inference...')
+            import tensorflow as tf
+            keras = False  # assume TF1 saved_model
+            model = tf.keras.models.load_model(w) if keras else tf.saved_model.load(w)
+        elif pb:  # GraphDef https://www.tensorflow.org/guide/migrate#a_graphpb_or_graphpbtxt
+            LOGGER.info(f'Loading {w} for TensorFlow GraphDef inference...')
+            import tensorflow as tf
+
+            def wrap_frozen_graph(gd, inputs, outputs):
+                x = tf.compat.v1.wrap_function(lambda: tf.compat.v1.import_graph_def(gd, name=''), [])  # wrapped
+                ge = x.graph.as_graph_element
+                return x.prune(tf.nest.map_structure(ge, inputs), tf.nest.map_structure(ge, outputs))
+
+            def gd_outputs(gd):
+                name_list, input_list = [], []
+                for node in gd.node:  # tensorflow.core.framework.node_def_pb2.NodeDef
+                    name_list.append(node.name)
+                    input_list.extend(node.input)
+                return sorted(f'{x}:0' for x in list(set(name_list) - set(input_list)) if not x.startswith('NoOp'))
+
+            gd = tf.Graph().as_graph_def()  # TF GraphDef
+            with open(w, 'rb') as f:
+                gd.ParseFromString(f.read())
+            frozen_func = wrap_frozen_graph(gd, inputs='x:0', outputs=gd_outputs(gd))
+        elif tflite or edgetpu:  # https://www.tensorflow.org/lite/guide/python#install_tensorflow_lite_for_python
+            try:  # https://coral.ai/docs/edgetpu/tflite-python/#update-existing-tf-lite-code-for-the-edge-tpu
+                from tflite_runtime.interpreter import Interpreter, load_delegate
+            except ImportError:
+                import tensorflow as tf
+                Interpreter, load_delegate = tf.lite.Interpreter, tf.lite.experimental.load_delegate,
+            if edgetpu:  # TF Edge TPU https://coral.ai/software/#edgetpu-runtime
+                LOGGER.info(f'Loading {w} for TensorFlow Lite Edge TPU inference...')
+                delegate = {
+                    'Linux': 'libedgetpu.so.1',
+                    'Darwin': 'libedgetpu.1.dylib',
+                    'Windows': 'edgetpu.dll'}[platform.system()]
+                interpreter = Interpreter(model_path=w, experimental_delegates=[load_delegate(delegate)])
+            else:  # TFLite
+                LOGGER.info(f'Loading {w} for TensorFlow Lite inference...')
+                interpreter = Interpreter(model_path=w)  # load TFLite model
+            interpreter.allocate_tensors()  # allocate
+            input_details = interpreter.get_input_details()  # inputs
+            output_details = interpreter.get_output_details()  # outputs
+            # load metadata
+            with contextlib.suppress(zipfile.BadZipFile):
+                with zipfile.ZipFile(w, 'r') as model:
+                    meta_file = model.namelist()[0]
+                    meta = ast.literal_eval(model.read(meta_file).decode('utf-8'))
+                    stride, names = int(meta['stride']), meta['names']
+        elif tfjs:  # TF.js
+            raise NotImplementedError('ERROR: YOLOv5 TF.js inference is not supported')
+        elif paddle:  # PaddlePaddle
+            LOGGER.info(f'Loading {w} for PaddlePaddle inference...')
+            check_requirements('paddlepaddle-gpu' if cuda else 'paddlepaddle')
+            import paddle.inference as pdi
+            if not Path(w).is_file():  # if not *.pdmodel
+                w = next(Path(w).rglob('*.pdmodel'))  # get *.pdmodel file from *_paddle_model dir
+            weights = Path(w).with_suffix('.pdiparams')
+            config = pdi.Config(str(w), str(weights))
+            if cuda:
+                config.enable_use_gpu(memory_pool_init_size_mb=2048, device_id=0)
+            predictor = pdi.create_predictor(config)
+            input_handle = predictor.get_input_handle(predictor.get_input_names()[0])
+            output_names = predictor.get_output_names()
+        elif triton:  # NVIDIA Triton Inference Server
+            LOGGER.info(f'Using {w} as Triton Inference Server...')
+            check_requirements('tritonclient[all]')
+            from utils.triton import TritonRemoteModel
+            model = TritonRemoteModel(url=w)
+            nhwc = model.runtime.startswith('tensorflow')
+        else:
+            raise NotImplementedError(f'ERROR: {w} is not a supported format')
+
+        # class names
+        if 'names' not in locals():
+            names = yaml_load(data)['names'] if data else {i: f'class{i}' for i in range(999)}
+        if names[0] == 'n01440764' and len(names) == 1000:  # ImageNet
+            names = yaml_load(ROOT / 'data/ImageNet.yaml')['names']  # human-readable names
+
+        self.__dict__.update(locals())  # assign all variables to self
+
+    def forward(self, im, augment=False, visualize=False):
+        # YOLOv5 MultiBackend inference
+        b, ch, h, w = im.shape  # batch, channel, height, width
+        if self.fp16 and im.dtype != torch.float16:
+            im = im.half()  # to FP16
+        if self.nhwc:
+            im = im.permute(0, 2, 3, 1)  # torch BCHW to numpy BHWC shape(1,320,192,3)
+
+        if self.pt:  # PyTorch
+            y = self.model(im, augment=augment, visualize=visualize) if augment or visualize else self.model(im)
+        elif self.jit:  # TorchScript
+            y = self.model(im)
+        elif self.dnn:  # ONNX OpenCV DNN
+            im = im.cpu().numpy()  # torch to numpy
+            self.net.setInput(im)
+            y = self.net.forward()
+        elif self.onnx:  # ONNX Runtime
+            im = im.cpu().numpy()  # torch to numpy
+            y = self.session.run(self.output_names, {self.session.get_inputs()[0].name: im})
+        elif self.xml:  # OpenVINO
+            im = im.cpu().numpy()  # FP32
+            y = list(self.ov_compiled_model(im).values())
+        elif self.engine:  # TensorRT
+            if self.dynamic and im.shape != self.bindings['images'].shape:
+                i = self.model.get_binding_index('images')
+                self.context.set_binding_shape(i, im.shape)  # reshape if dynamic
+                self.bindings['images'] = self.bindings['images']._replace(shape=im.shape)
+                for name in self.output_names:
+                    i = self.model.get_binding_index(name)
+                    self.bindings[name].data.resize_(tuple(self.context.get_binding_shape(i)))
+            s = self.bindings['images'].shape
+            assert im.shape == s, f"input size {im.shape} {'>' if self.dynamic else 'not equal to'} max model size {s}"
+            self.binding_addrs['images'] = int(im.data_ptr())
+            self.context.execute_v2(list(self.binding_addrs.values()))
+            y = [self.bindings[x].data for x in sorted(self.output_names)]
+        elif self.coreml:  # CoreML
+            im = im.cpu().numpy()
+            im = Image.fromarray((im[0] * 255).astype('uint8'))
+            # im = im.resize((192, 320), Image.BILINEAR)
+            y = self.model.predict({'image': im})  # coordinates are xywh normalized
+            if 'confidence' in y:
+                box = xywh2xyxy(y['coordinates'] * [[w, h, w, h]])  # xyxy pixels
+                conf, cls = y['confidence'].max(1), y['confidence'].argmax(1).astype(np.float)
+                y = np.concatenate((box, conf.reshape(-1, 1), cls.reshape(-1, 1)), 1)
+            else:
+                y = list(reversed(y.values()))  # reversed for segmentation models (pred, proto)
+        elif self.paddle:  # PaddlePaddle
+            im = im.cpu().numpy().astype(np.float32)
+            self.input_handle.copy_from_cpu(im)
+            self.predictor.run()
+            y = [self.predictor.get_output_handle(x).copy_to_cpu() for x in self.output_names]
+        elif self.triton:  # NVIDIA Triton Inference Server
+            y = self.model(im)
+        else:  # TensorFlow (SavedModel, GraphDef, Lite, Edge TPU)
+            im = im.cpu().numpy()
+            if self.saved_model:  # SavedModel
+                y = self.model(im, training=False) if self.keras else self.model(im)
+            elif self.pb:  # GraphDef
+                y = self.frozen_func(x=self.tf.constant(im))
+            else:  # Lite or Edge TPU
+                input = self.input_details[0]
+                int8 = input['dtype'] == np.uint8  # is TFLite quantized uint8 model
+                if int8:
+                    scale, zero_point = input['quantization']
+                    im = (im / scale + zero_point).astype(np.uint8)  # de-scale
+                self.interpreter.set_tensor(input['index'], im)
+                self.interpreter.invoke()
+                y = []
+                for output in self.output_details:
+                    x = self.interpreter.get_tensor(output['index'])
+                    if int8:
+                        scale, zero_point = output['quantization']
+                        x = (x.astype(np.float32) - zero_point) * scale  # re-scale
+                    y.append(x)
+            y = [x if isinstance(x, np.ndarray) else x.numpy() for x in y]
+            y[0][..., :4] *= [w, h, w, h]  # xywh normalized to pixels
+
+        if isinstance(y, (list, tuple)):
+            return self.from_numpy(y[0]) if len(y) == 1 else [self.from_numpy(x) for x in y]
+        else:
+            return self.from_numpy(y)
+
+    def from_numpy(self, x):
+        return torch.from_numpy(x).to(self.device) if isinstance(x, np.ndarray) else x
+
+    def warmup(self, imgsz=(1, 3, 640, 640)):
+        # Warmup model by running inference once
+        warmup_types = self.pt, self.jit, self.onnx, self.engine, self.saved_model, self.pb, self.triton
+        if any(warmup_types) and (self.device.type != 'cpu' or self.triton):
+            im = torch.empty(*imgsz, dtype=torch.half if self.fp16 else torch.float, device=self.device)  # input
+            for _ in range(2 if self.jit else 1):  #
+                self.forward(im)  # warmup
+
+    @staticmethod
+    def _model_type(p='path/to/model.pt'):
+        # Return model type from model path, i.e. path='path/to/model.onnx' -> type=onnx
+        # types = [pt, jit, onnx, xml, engine, coreml, saved_model, pb, tflite, edgetpu, tfjs, paddle]
+        from export import export_formats
+        from utils.downloads import is_url
+        sf = list(export_formats().Suffix)  # export suffixes
+        if not is_url(p, check=False):
+            check_suffix(p, sf)  # checks
+        url = urlparse(p)  # if url may be Triton inference server
+        types = [s in Path(p).name for s in sf]
+        types[8] &= not types[9]  # tflite &= not edgetpu
+        triton = not any(types) and all([any(s in url.scheme for s in ['http', 'grpc']), url.netloc])
+        return types + [triton]
+
+    @staticmethod
+    def _load_metadata(f=Path('path/to/meta.yaml')):
+        # Load metadata from meta.yaml if it exists
+        if f.exists():
+            d = yaml_load(f)
+            return d['stride'], d['names']  # assign stride, names
+        return None, None
+
+
+class AutoShape(nn.Module):
+    # YOLOv5 input-robust model wrapper for passing cv2/np/PIL/torch inputs. Includes preprocessing, inference and NMS
+    conf = 0.25  # NMS confidence threshold
+    iou = 0.45  # NMS IoU threshold
+    agnostic = False  # NMS class-agnostic
+    multi_label = False  # NMS multiple labels per box
+    classes = None  # (optional list) filter by class, i.e. = [0, 15, 16] for COCO persons, cats and dogs
+    max_det = 1000  # maximum number of detections per image
+    amp = False  # Automatic Mixed Precision (AMP) inference
+
+    def __init__(self, model, verbose=True):
+        super().__init__()
+        if verbose:
+            LOGGER.info('Adding AutoShape... ')
+        copy_attr(self, model, include=('yaml', 'nc', 'hyp', 'names', 'stride', 'abc'), exclude=())  # copy attributes
+        self.dmb = isinstance(model, DetectMultiBackend)  # DetectMultiBackend() instance
+        self.pt = not self.dmb or model.pt  # PyTorch model
+        self.model = model.eval()
+        if self.pt:
+            m = self.model.model.model[-1] if self.dmb else self.model.model[-1]  # Detect()
+            m.inplace = False  # Detect.inplace=False for safe multithread inference
+            m.export = True  # do not output loss values
+
+    def _apply(self, fn):
+        # Apply to(), cpu(), cuda(), half() to model tensors that are not parameters or registered buffers
+        self = super()._apply(fn)
+        if self.pt:
+            m = self.model.model.model[-1] if self.dmb else self.model.model[-1]  # Detect()
+            m.stride = fn(m.stride)
+            m.grid = list(map(fn, m.grid))
+            if isinstance(m.anchor_grid, list):
+                m.anchor_grid = list(map(fn, m.anchor_grid))
+        return self
+
+    @smart_inference_mode()
+    def forward(self, ims, size=640, augment=False, profile=False):
+        # Inference from various sources. For size(height=640, width=1280), RGB images example inputs are:
+        #   file:        ims = 'data/images/zidane.jpg'  # str or PosixPath
+        #   URI:             = 'https://ultralytics.com/images/zidane.jpg'
+        #   OpenCV:          = cv2.imread('image.jpg')[:,:,::-1]  # HWC BGR to RGB x(640,1280,3)
+        #   PIL:             = Image.open('image.jpg') or ImageGrab.grab()  # HWC x(640,1280,3)
+        #   numpy:           = np.zeros((640,1280,3))  # HWC
+        #   torch:           = torch.zeros(16,3,320,640)  # BCHW (scaled to size=640, 0-1 values)
+        #   multiple:        = [Image.open('image1.jpg'), Image.open('image2.jpg'), ...]  # list of images
+
+        dt = (Profile(), Profile(), Profile())
+        with dt[0]:
+            if isinstance(size, int):  # expand
+                size = (size, size)
+            p = next(self.model.parameters()) if self.pt else torch.empty(1, device=self.model.device)  # param
+            autocast = self.amp and (p.device.type != 'cpu')  # Automatic Mixed Precision (AMP) inference
+            if isinstance(ims, torch.Tensor):  # torch
+                with amp.autocast(autocast):
+                    return self.model(ims.to(p.device).type_as(p), augment=augment)  # inference
+
+            # Pre-process
+            n, ims = (len(ims), list(ims)) if isinstance(ims, (list, tuple)) else (1, [ims])  # number, list of images
+            shape0, shape1, files = [], [], []  # image and inference shapes, filenames
+            for i, im in enumerate(ims):
+                f = f'image{i}'  # filename
+                if isinstance(im, (str, Path)):  # filename or uri
+                    im, f = Image.open(requests.get(im, stream=True).raw if str(im).startswith('http') else im), im
+                    im = np.asarray(exif_transpose(im))
+                elif isinstance(im, Image.Image):  # PIL Image
+                    im, f = np.asarray(exif_transpose(im)), getattr(im, 'filename', f) or f
+                files.append(Path(f).with_suffix('.jpg').name)
+                if im.shape[0] < 5:  # image in CHW
+                    im = im.transpose((1, 2, 0))  # reverse dataloader .transpose(2, 0, 1)
+                im = im[..., :3] if im.ndim == 3 else cv2.cvtColor(im, cv2.COLOR_GRAY2BGR)  # enforce 3ch input
+                s = im.shape[:2]  # HWC
+                shape0.append(s)  # image shape
+                g = max(size) / max(s)  # gain
+                shape1.append([int(y * g) for y in s])
+                ims[i] = im if im.data.contiguous else np.ascontiguousarray(im)  # update
+            shape1 = [make_divisible(x, self.stride) for x in np.array(shape1).max(0)]  # inf shape
+            x = [letterbox(im, shape1, auto=False)[0] for im in ims]  # pad
+            x = np.ascontiguousarray(np.array(x).transpose((0, 3, 1, 2)))  # stack and BHWC to BCHW
+            x = torch.from_numpy(x).to(p.device).type_as(p) / 255  # uint8 to fp16/32
+
+        with amp.autocast(autocast):
+            # Inference
+            with dt[1]:
+                y = self.model(x, augment=augment)  # forward
+
+            # Post-process
+            with dt[2]:
+                y = non_max_suppression(y if self.dmb else y[0],
+                                        self.conf,
+                                        self.iou,
+                                        self.classes,
+                                        self.agnostic,
+                                        self.multi_label,
+                                        max_det=self.max_det)  # NMS
+                for i in range(n):
+                    scale_boxes(shape1, y[i][:, :4], shape0[i])
+
+            return Detections(ims, y, files, dt, self.names, x.shape)
+
+
+class Detections:
+    # YOLOv5 detections class for inference results
+    def __init__(self, ims, pred, files, times=(0, 0, 0), names=None, shape=None):
+        super().__init__()
+        d = pred[0].device  # device
+        gn = [torch.tensor([*(im.shape[i] for i in [1, 0, 1, 0]), 1, 1], device=d) for im in ims]  # normalizations
+        self.ims = ims  # list of images as numpy arrays
+        self.pred = pred  # list of tensors pred[0] = (xyxy, conf, cls)
+        self.names = names  # class names
+        self.files = files  # image filenames
+        self.times = times  # profiling times
+        self.xyxy = pred  # xyxy pixels
+        self.xywh = [xyxy2xywh(x) for x in pred]  # xywh pixels
+        self.xyxyn = [x / g for x, g in zip(self.xyxy, gn)]  # xyxy normalized
+        self.xywhn = [x / g for x, g in zip(self.xywh, gn)]  # xywh normalized
+        self.n = len(self.pred)  # number of images (batch size)
+        self.t = tuple(x.t / self.n * 1E3 for x in times)  # timestamps (ms)
+        self.s = tuple(shape)  # inference BCHW shape
+
+    def _run(self, pprint=False, show=False, save=False, crop=False, render=False, labels=True, save_dir=Path('')):
+        s, crops = '', []
+        for i, (im, pred) in enumerate(zip(self.ims, self.pred)):
+            s += f'\nimage {i + 1}/{len(self.pred)}: {im.shape[0]}x{im.shape[1]} '  # string
+            if pred.shape[0]:
+                for c in pred[:, -1].unique():
+                    n = (pred[:, -1] == c).sum()  # detections per class
+                    s += f"{n} {self.names[int(c)]}{'s' * (n > 1)}, "  # add to string
+                s = s.rstrip(', ')
+                if show or save or render or crop:
+                    annotator = Annotator(im, example=str(self.names))
+                    for *box, conf, cls in reversed(pred):  # xyxy, confidence, class
+                        label = f'{self.names[int(cls)]} {conf:.2f}'
+                        if crop:
+                            file = save_dir / 'crops' / self.names[int(cls)] / self.files[i] if save else None
+                            crops.append({
+                                'box': box,
+                                'conf': conf,
+                                'cls': cls,
+                                'label': label,
+                                'im': save_one_box(box, im, file=file, save=save)})
+                        else:  # all others
+                            annotator.box_label(box, label if labels else '', color=colors(cls))
+                    im = annotator.im
+            else:
+                s += '(no detections)'
+
+            im = Image.fromarray(im.astype(np.uint8)) if isinstance(im, np.ndarray) else im  # from np
+            if show:
+                if is_jupyter():
+                    from IPython.display import display
+                    display(im)
+                else:
+                    im.show(self.files[i])
+            if save:
+                f = self.files[i]
+                im.save(save_dir / f)  # save
+                if i == self.n - 1:
+                    LOGGER.info(f"Saved {self.n} image{'s' * (self.n > 1)} to {colorstr('bold', save_dir)}")
+            if render:
+                self.ims[i] = np.asarray(im)
+        if pprint:
+            s = s.lstrip('\n')
+            return f'{s}\nSpeed: %.1fms pre-process, %.1fms inference, %.1fms NMS per image at shape {self.s}' % self.t
+        if crop:
+            if save:
+                LOGGER.info(f'Saved results to {save_dir}\n')
+            return crops
+
+    @TryExcept('Showing images is not supported in this environment')
+    def show(self, labels=True):
+        self._run(show=True, labels=labels)  # show results
+
+    def save(self, labels=True, save_dir='runs/detect/exp', exist_ok=False):
+        save_dir = increment_path(save_dir, exist_ok, mkdir=True)  # increment save_dir
+        self._run(save=True, labels=labels, save_dir=save_dir)  # save results
+
+    def crop(self, save=True, save_dir='runs/detect/exp', exist_ok=False):
+        save_dir = increment_path(save_dir, exist_ok, mkdir=True) if save else None
+        return self._run(crop=True, save=save, save_dir=save_dir)  # crop results
+
+    def render(self, labels=True):
+        self._run(render=True, labels=labels)  # render results
+        return self.ims
+
+    def pandas(self):
+        # return detections as pandas DataFrames, i.e. print(results.pandas().xyxy[0])
+        new = copy(self)  # return copy
+        ca = 'xmin', 'ymin', 'xmax', 'ymax', 'confidence', 'class', 'name'  # xyxy columns
+        cb = 'xcenter', 'ycenter', 'width', 'height', 'confidence', 'class', 'name'  # xywh columns
+        for k, c in zip(['xyxy', 'xyxyn', 'xywh', 'xywhn'], [ca, ca, cb, cb]):
+            a = [[x[:5] + [int(x[5]), self.names[int(x[5])]] for x in x.tolist()] for x in getattr(self, k)]  # update
+            setattr(new, k, [pd.DataFrame(x, columns=c) for x in a])
+        return new
+
+    def tolist(self):
+        # return a list of Detections objects, i.e. 'for result in results.tolist():'
+        r = range(self.n)  # iterable
+        x = [Detections([self.ims[i]], [self.pred[i]], [self.files[i]], self.times, self.names, self.s) for i in r]
+        # for d in x:
+        #    for k in ['ims', 'pred', 'xyxy', 'xyxyn', 'xywh', 'xywhn']:
+        #        setattr(d, k, getattr(d, k)[0])  # pop out of list
+        return x
+
+    def print(self):
+        LOGGER.info(self.__str__())
+
+    def __len__(self):  # override len(results)
+        return self.n
+
+    def __str__(self):  # override print(results)
+        return self._run(pprint=True)  # print results
+
+    def __repr__(self):
+        return f'YOLOv5 {self.__class__} instance\n' + self.__str__()
+
+
+class Proto(nn.Module):
+    # YOLOv5 mask Proto module for segmentation models
+    def __init__(self, c1, c_=256, c2=32):  # ch_in, number of protos, number of masks
+        super().__init__()
+        self.cv1 = Conv(c1, c_, k=3)
+        self.upsample = nn.Upsample(scale_factor=2, mode='nearest')
+        self.cv2 = Conv(c_, c_, k=3)
+        self.cv3 = Conv(c_, c2)
+
+    def forward(self, x):
+        return self.cv3(self.cv2(self.upsample(self.cv1(x))))
+
+
+class Classify(nn.Module):
+    # YOLOv5 classification head, i.e. x(b,c1,20,20) to x(b,c2)
+    def __init__(self,
+                 c1,
+                 c2,
+                 k=1,
+                 s=1,
+                 p=None,
+                 g=1,
+                 dropout_p=0.0):  # ch_in, ch_out, kernel, stride, padding, groups, dropout probability
+        super().__init__()
+        c_ = 1280  # efficientnet_b0 size
+        self.conv = Conv(c1, c_, k, s, autopad(k, p), g)
+        self.pool = nn.AdaptiveAvgPool2d(1)  # to x(b,c_,1,1)
+        self.drop = nn.Dropout(p=dropout_p, inplace=True)
+        self.linear = nn.Linear(c_, c2)  # to x(b,c2)
+
+    def forward(self, x):
+        if isinstance(x, list):
+            x = torch.cat(x, 1)
+        return self.linear(self.drop(self.pool(self.conv(x)).flatten(1)))

models/experimental.py

@@ -0,0 +1,111 @@
+# YOLOv5 🚀 by Ultralytics, AGPL-3.0 license
+"""
+Experimental modules
+"""
+import math
+
+import numpy as np
+import torch
+import torch.nn as nn
+
+from utils.downloads import attempt_download
+
+
+class Sum(nn.Module):
+    # Weighted sum of 2 or more layers https://arxiv.org/abs/1911.09070
+    def __init__(self, n, weight=False):  # n: number of inputs
+        super().__init__()
+        self.weight = weight  # apply weights boolean
+        self.iter = range(n - 1)  # iter object
+        if weight:
+            self.w = nn.Parameter(-torch.arange(1.0, n) / 2, requires_grad=True)  # layer weights
+
+    def forward(self, x):
+        y = x[0]  # no weight
+        if self.weight:
+            w = torch.sigmoid(self.w) * 2
+            for i in self.iter:
+                y = y + x[i + 1] * w[i]
+        else:
+            for i in self.iter:
+                y = y + x[i + 1]
+        return y
+
+
+class MixConv2d(nn.Module):
+    # Mixed Depth-wise Conv https://arxiv.org/abs/1907.09595
+    def __init__(self, c1, c2, k=(1, 3), s=1, equal_ch=True):  # ch_in, ch_out, kernel, stride, ch_strategy
+        super().__init__()
+        n = len(k)  # number of convolutions
+        if equal_ch:  # equal c_ per group
+            i = torch.linspace(0, n - 1E-6, c2).floor()  # c2 indices
+            c_ = [(i == g).sum() for g in range(n)]  # intermediate channels
+        else:  # equal weight.numel() per group
+            b = [c2] + [0] * n
+            a = np.eye(n + 1, n, k=-1)
+            a -= np.roll(a, 1, axis=1)
+            a *= np.array(k) ** 2
+            a[0] = 1
+            c_ = np.linalg.lstsq(a, b, rcond=None)[0].round()  # solve for equal weight indices, ax = b
+
+        self.m = nn.ModuleList([
+            nn.Conv2d(c1, int(c_), k, s, k // 2, groups=math.gcd(c1, int(c_)), bias=False) for k, c_ in zip(k, c_)])
+        self.bn = nn.BatchNorm2d(c2)
+        self.act = nn.SiLU()
+
+    def forward(self, x):
+        return self.act(self.bn(torch.cat([m(x) for m in self.m], 1)))
+
+
+class Ensemble(nn.ModuleList):
+    # Ensemble of models
+    def __init__(self):
+        super().__init__()
+
+    def forward(self, x, augment=False, profile=False, visualize=False):
+        y = [module(x, augment, profile, visualize)[0] for module in self]
+        # y = torch.stack(y).max(0)[0]  # max ensemble
+        # y = torch.stack(y).mean(0)  # mean ensemble
+        y = torch.cat(y, 1)  # nms ensemble
+        return y, None  # inference, train output
+
+
+def attempt_load(weights, device=None, inplace=True, fuse=True):
+    # Loads an ensemble of models weights=[a,b,c] or a single model weights=[a] or weights=a
+    from models.yolo import Detect, Model
+
+    model = Ensemble()
+    for w in weights if isinstance(weights, list) else [weights]:
+        ckpt = torch.load(attempt_download(w), map_location='cpu')  # load
+        ckpt = (ckpt.get('ema') or ckpt['model']).to(device).float()  # FP32 model
+
+        # Model compatibility updates
+        if not hasattr(ckpt, 'stride'):
+            ckpt.stride = torch.tensor([32.])
+        if hasattr(ckpt, 'names') and isinstance(ckpt.names, (list, tuple)):
+            ckpt.names = dict(enumerate(ckpt.names))  # convert to dict
+
+        model.append(ckpt.fuse().eval() if fuse and hasattr(ckpt, 'fuse') else ckpt.eval())  # model in eval mode
+
+    # Module updates
+    for m in model.modules():
+        t = type(m)
+        if t in (nn.Hardswish, nn.LeakyReLU, nn.ReLU, nn.ReLU6, nn.SiLU, Detect, Model):
+            m.inplace = inplace
+            if t is Detect and not isinstance(m.anchor_grid, list):
+                delattr(m, 'anchor_grid')
+                setattr(m, 'anchor_grid', [torch.zeros(1)] * m.nl)
+        elif t is nn.Upsample and not hasattr(m, 'recompute_scale_factor'):
+            m.recompute_scale_factor = None  # torch 1.11.0 compatibility
+
+    # Return model
+    if len(model) == 1:
+        return model[-1]
+
+    # Return detection ensemble
+    print(f'Ensemble created with {weights}\n')
+    for k in 'names', 'nc', 'yaml':
+        setattr(model, k, getattr(model[0], k))
+    model.stride = model[torch.argmax(torch.tensor([m.stride.max() for m in model])).int()].stride  # max stride
+    assert all(model[0].nc == m.nc for m in model), f'Models have different class counts: {[m.nc for m in model]}'
+    return model

models/hub/anchors.yaml

@@ -0,0 +1,59 @@
+# YOLOv5 🚀 by Ultralytics, AGPL-3.0 license
+# Default anchors for COCO data
+
+
+# P5 -------------------------------------------------------------------------------------------------------------------
+# P5-640:
+anchors_p5_640:
+  - [10,13, 16,30, 33,23]  # P3/8
+  - [30,61, 62,45, 59,119]  # P4/16
+  - [116,90, 156,198, 373,326]  # P5/32
+
+
+# P6 -------------------------------------------------------------------------------------------------------------------
+# P6-640:  thr=0.25: 0.9964 BPR, 5.54 anchors past thr, n=12, img_size=640, metric_all=0.281/0.716-mean/best, past_thr=0.469-mean: 9,11,  21,19,  17,41,  43,32,  39,70,  86,64,  65,131,  134,130,  120,265,  282,180,  247,354,  512,387
+anchors_p6_640:
+  - [9,11,  21,19,  17,41]  # P3/8
+  - [43,32,  39,70,  86,64]  # P4/16
+  - [65,131,  134,130,  120,265]  # P5/32
+  - [282,180,  247,354,  512,387]  # P6/64
+
+# P6-1280:  thr=0.25: 0.9950 BPR, 5.55 anchors past thr, n=12, img_size=1280, metric_all=0.281/0.714-mean/best, past_thr=0.468-mean: 19,27,  44,40,  38,94,  96,68,  86,152,  180,137,  140,301,  303,264,  238,542,  436,615,  739,380,  925,792
+anchors_p6_1280:
+  - [19,27,  44,40,  38,94]  # P3/8
+  - [96,68,  86,152,  180,137]  # P4/16
+  - [140,301,  303,264,  238,542]  # P5/32
+  - [436,615,  739,380,  925,792]  # P6/64
+
+# P6-1920:  thr=0.25: 0.9950 BPR, 5.55 anchors past thr, n=12, img_size=1920, metric_all=0.281/0.714-mean/best, past_thr=0.468-mean: 28,41,  67,59,  57,141,  144,103,  129,227,  270,205,  209,452,  455,396,  358,812,  653,922,  1109,570,  1387,1187
+anchors_p6_1920:
+  - [28,41,  67,59,  57,141]  # P3/8
+  - [144,103,  129,227,  270,205]  # P4/16
+  - [209,452,  455,396,  358,812]  # P5/32
+  - [653,922,  1109,570,  1387,1187]  # P6/64
+
+
+# P7 -------------------------------------------------------------------------------------------------------------------
+# P7-640:  thr=0.25: 0.9962 BPR, 6.76 anchors past thr, n=15, img_size=640, metric_all=0.275/0.733-mean/best, past_thr=0.466-mean: 11,11,  13,30,  29,20,  30,46,  61,38,  39,92,  78,80,  146,66,  79,163,  149,150,  321,143,  157,303,  257,402,  359,290,  524,372
+anchors_p7_640:
+  - [11,11,  13,30,  29,20]  # P3/8
+  - [30,46,  61,38,  39,92]  # P4/16
+  - [78,80,  146,66,  79,163]  # P5/32
+  - [149,150,  321,143,  157,303]  # P6/64
+  - [257,402,  359,290,  524,372]  # P7/128
+
+# P7-1280:  thr=0.25: 0.9968 BPR, 6.71 anchors past thr, n=15, img_size=1280, metric_all=0.273/0.732-mean/best, past_thr=0.463-mean: 19,22,  54,36,  32,77,  70,83,  138,71,  75,173,  165,159,  148,334,  375,151,  334,317,  251,626,  499,474,  750,326,  534,814,  1079,818
+anchors_p7_1280:
+  - [19,22,  54,36,  32,77]  # P3/8
+  - [70,83,  138,71,  75,173]  # P4/16
+  - [165,159,  148,334,  375,151]  # P5/32
+  - [334,317,  251,626,  499,474]  # P6/64
+  - [750,326,  534,814,  1079,818]  # P7/128
+
+# P7-1920:  thr=0.25: 0.9968 BPR, 6.71 anchors past thr, n=15, img_size=1920, metric_all=0.273/0.732-mean/best, past_thr=0.463-mean: 29,34,  81,55,  47,115,  105,124,  207,107,  113,259,  247,238,  222,500,  563,227,  501,476,  376,939,  749,711,  1126,489,  801,1222,  1618,1227
+anchors_p7_1920:
+  - [29,34,  81,55,  47,115]  # P3/8
+  - [105,124,  207,107,  113,259]  # P4/16
+  - [247,238,  222,500,  563,227]  # P5/32
+  - [501,476,  376,939,  749,711]  # P6/64
+  - [1126,489,  801,1222,  1618,1227]  # P7/128

models/hub/yolov3-spp.yaml

@@ -0,0 +1,51 @@
+# YOLOv5 🚀 by Ultralytics, AGPL-3.0 license
+
+# Parameters
+nc: 80  # number of classes
+depth_multiple: 1.0  # model depth multiple
+width_multiple: 1.0  # layer channel multiple
+anchors:
+  - [10,13, 16,30, 33,23]  # P3/8
+  - [30,61, 62,45, 59,119]  # P4/16
+  - [116,90, 156,198, 373,326]  # P5/32
+
+# darknet53 backbone
+backbone:
+  # [from, number, module, args]
+  [[-1, 1, Conv, [32, 3, 1]],  # 0
+   [-1, 1, Conv, [64, 3, 2]],  # 1-P1/2
+   [-1, 1, Bottleneck, [64]],
+   [-1, 1, Conv, [128, 3, 2]],  # 3-P2/4
+   [-1, 2, Bottleneck, [128]],
+   [-1, 1, Conv, [256, 3, 2]],  # 5-P3/8
+   [-1, 8, Bottleneck, [256]],
+   [-1, 1, Conv, [512, 3, 2]],  # 7-P4/16
+   [-1, 8, Bottleneck, [512]],
+   [-1, 1, Conv, [1024, 3, 2]],  # 9-P5/32
+   [-1, 4, Bottleneck, [1024]],  # 10
+  ]
+
+# YOLOv3-SPP head
+head:
+  [[-1, 1, Bottleneck, [1024, False]],
+   [-1, 1, SPP, [512, [5, 9, 13]]],
+   [-1, 1, Conv, [1024, 3, 1]],
+   [-1, 1, Conv, [512, 1, 1]],
+   [-1, 1, Conv, [1024, 3, 1]],  # 15 (P5/32-large)
+
+   [-2, 1, Conv, [256, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 8], 1, Concat, [1]],  # cat backbone P4
+   [-1, 1, Bottleneck, [512, False]],
+   [-1, 1, Bottleneck, [512, False]],
+   [-1, 1, Conv, [256, 1, 1]],
+   [-1, 1, Conv, [512, 3, 1]],  # 22 (P4/16-medium)
+
+   [-2, 1, Conv, [128, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 6], 1, Concat, [1]],  # cat backbone P3
+   [-1, 1, Bottleneck, [256, False]],
+   [-1, 2, Bottleneck, [256, False]],  # 27 (P3/8-small)
+
+   [[27, 22, 15], 1, Detect, [nc, anchors]],   # Detect(P3, P4, P5)
+  ]

models/hub/yolov3-tiny.yaml

@@ -0,0 +1,41 @@
+# YOLOv5 🚀 by Ultralytics, AGPL-3.0 license
+
+# Parameters
+nc: 80  # number of classes
+depth_multiple: 1.0  # model depth multiple
+width_multiple: 1.0  # layer channel multiple
+anchors:
+  - [10,14, 23,27, 37,58]  # P4/16
+  - [81,82, 135,169, 344,319]  # P5/32
+
+# YOLOv3-tiny backbone
+backbone:
+  # [from, number, module, args]
+  [[-1, 1, Conv, [16, 3, 1]],  # 0
+   [-1, 1, nn.MaxPool2d, [2, 2, 0]],  # 1-P1/2
+   [-1, 1, Conv, [32, 3, 1]],
+   [-1, 1, nn.MaxPool2d, [2, 2, 0]],  # 3-P2/4
+   [-1, 1, Conv, [64, 3, 1]],
+   [-1, 1, nn.MaxPool2d, [2, 2, 0]],  # 5-P3/8
+   [-1, 1, Conv, [128, 3, 1]],
+   [-1, 1, nn.MaxPool2d, [2, 2, 0]],  # 7-P4/16
+   [-1, 1, Conv, [256, 3, 1]],
+   [-1, 1, nn.MaxPool2d, [2, 2, 0]],  # 9-P5/32
+   [-1, 1, Conv, [512, 3, 1]],
+   [-1, 1, nn.ZeroPad2d, [[0, 1, 0, 1]]],  # 11
+   [-1, 1, nn.MaxPool2d, [2, 1, 0]],  # 12
+  ]
+
+# YOLOv3-tiny head
+head:
+  [[-1, 1, Conv, [1024, 3, 1]],
+   [-1, 1, Conv, [256, 1, 1]],
+   [-1, 1, Conv, [512, 3, 1]],  # 15 (P5/32-large)
+
+   [-2, 1, Conv, [128, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 8], 1, Concat, [1]],  # cat backbone P4
+   [-1, 1, Conv, [256, 3, 1]],  # 19 (P4/16-medium)
+
+   [[19, 15], 1, Detect, [nc, anchors]],  # Detect(P4, P5)
+  ]

models/hub/yolov3.yaml

@@ -0,0 +1,51 @@
+# YOLOv5 🚀 by Ultralytics, AGPL-3.0 license
+
+# Parameters
+nc: 80  # number of classes
+depth_multiple: 1.0  # model depth multiple
+width_multiple: 1.0  # layer channel multiple
+anchors:
+  - [10,13, 16,30, 33,23]  # P3/8
+  - [30,61, 62,45, 59,119]  # P4/16
+  - [116,90, 156,198, 373,326]  # P5/32
+
+# darknet53 backbone
+backbone:
+  # [from, number, module, args]
+  [[-1, 1, Conv, [32, 3, 1]],  # 0
+   [-1, 1, Conv, [64, 3, 2]],  # 1-P1/2
+   [-1, 1, Bottleneck, [64]],
+   [-1, 1, Conv, [128, 3, 2]],  # 3-P2/4
+   [-1, 2, Bottleneck, [128]],
+   [-1, 1, Conv, [256, 3, 2]],  # 5-P3/8
+   [-1, 8, Bottleneck, [256]],
+   [-1, 1, Conv, [512, 3, 2]],  # 7-P4/16
+   [-1, 8, Bottleneck, [512]],
+   [-1, 1, Conv, [1024, 3, 2]],  # 9-P5/32
+   [-1, 4, Bottleneck, [1024]],  # 10
+  ]
+
+# YOLOv3 head
+head:
+  [[-1, 1, Bottleneck, [1024, False]],
+   [-1, 1, Conv, [512, 1, 1]],
+   [-1, 1, Conv, [1024, 3, 1]],
+   [-1, 1, Conv, [512, 1, 1]],
+   [-1, 1, Conv, [1024, 3, 1]],  # 15 (P5/32-large)
+
+   [-2, 1, Conv, [256, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 8], 1, Concat, [1]],  # cat backbone P4
+   [-1, 1, Bottleneck, [512, False]],
+   [-1, 1, Bottleneck, [512, False]],
+   [-1, 1, Conv, [256, 1, 1]],
+   [-1, 1, Conv, [512, 3, 1]],  # 22 (P4/16-medium)
+
+   [-2, 1, Conv, [128, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 6], 1, Concat, [1]],  # cat backbone P3
+   [-1, 1, Bottleneck, [256, False]],
+   [-1, 2, Bottleneck, [256, False]],  # 27 (P3/8-small)
+
+   [[27, 22, 15], 1, Detect, [nc, anchors]],   # Detect(P3, P4, P5)
+  ]

models/hub/yolov5-bifpn.yaml

@@ -0,0 +1,48 @@
+# YOLOv5 🚀 by Ultralytics, AGPL-3.0 license
+
+# Parameters
+nc: 80  # number of classes
+depth_multiple: 1.0  # model depth multiple
+width_multiple: 1.0  # layer channel multiple
+anchors:
+  - [10,13, 16,30, 33,23]  # P3/8
+  - [30,61, 62,45, 59,119]  # P4/16
+  - [116,90, 156,198, 373,326]  # P5/32
+
+# YOLOv5 v6.0 backbone
+backbone:
+  # [from, number, module, args]
+  [[-1, 1, Conv, [64, 6, 2, 2]],  # 0-P1/2
+   [-1, 1, Conv, [128, 3, 2]],  # 1-P2/4
+   [-1, 3, C3, [128]],
+   [-1, 1, Conv, [256, 3, 2]],  # 3-P3/8
+   [-1, 6, C3, [256]],
+   [-1, 1, Conv, [512, 3, 2]],  # 5-P4/16
+   [-1, 9, C3, [512]],
+   [-1, 1, Conv, [1024, 3, 2]],  # 7-P5/32
+   [-1, 3, C3, [1024]],
+   [-1, 1, SPPF, [1024, 5]],  # 9
+  ]
+
+# YOLOv5 v6.0 BiFPN head
+head:
+  [[-1, 1, Conv, [512, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 6], 1, Concat, [1]],  # cat backbone P4
+   [-1, 3, C3, [512, False]],  # 13
+
+   [-1, 1, Conv, [256, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 4], 1, Concat, [1]],  # cat backbone P3
+   [-1, 3, C3, [256, False]],  # 17 (P3/8-small)
+
+   [-1, 1, Conv, [256, 3, 2]],
+   [[-1, 14, 6], 1, Concat, [1]],  # cat P4 <--- BiFPN change
+   [-1, 3, C3, [512, False]],  # 20 (P4/16-medium)
+
+   [-1, 1, Conv, [512, 3, 2]],
+   [[-1, 10], 1, Concat, [1]],  # cat head P5
+   [-1, 3, C3, [1024, False]],  # 23 (P5/32-large)
+
+   [[17, 20, 23], 1, Detect, [nc, anchors]],  # Detect(P3, P4, P5)
+  ]

models/hub/yolov5-fpn.yaml

@@ -0,0 +1,42 @@
+# YOLOv5 🚀 by Ultralytics, AGPL-3.0 license
+
+# Parameters
+nc: 80  # number of classes
+depth_multiple: 1.0  # model depth multiple
+width_multiple: 1.0  # layer channel multiple
+anchors:
+  - [10,13, 16,30, 33,23]  # P3/8
+  - [30,61, 62,45, 59,119]  # P4/16
+  - [116,90, 156,198, 373,326]  # P5/32
+
+# YOLOv5 v6.0 backbone
+backbone:
+  # [from, number, module, args]
+  [[-1, 1, Conv, [64, 6, 2, 2]],  # 0-P1/2
+   [-1, 1, Conv, [128, 3, 2]],  # 1-P2/4
+   [-1, 3, C3, [128]],
+   [-1, 1, Conv, [256, 3, 2]],  # 3-P3/8
+   [-1, 6, C3, [256]],
+   [-1, 1, Conv, [512, 3, 2]],  # 5-P4/16
+   [-1, 9, C3, [512]],
+   [-1, 1, Conv, [1024, 3, 2]],  # 7-P5/32
+   [-1, 3, C3, [1024]],
+   [-1, 1, SPPF, [1024, 5]],  # 9
+  ]
+
+# YOLOv5 v6.0 FPN head
+head:
+  [[-1, 3, C3, [1024, False]],  # 10 (P5/32-large)
+
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 6], 1, Concat, [1]],  # cat backbone P4
+   [-1, 1, Conv, [512, 1, 1]],
+   [-1, 3, C3, [512, False]],  # 14 (P4/16-medium)
+
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 4], 1, Concat, [1]],  # cat backbone P3
+   [-1, 1, Conv, [256, 1, 1]],
+   [-1, 3, C3, [256, False]],  # 18 (P3/8-small)
+
+   [[18, 14, 10], 1, Detect, [nc, anchors]],  # Detect(P3, P4, P5)
+  ]

models/hub/yolov5-p2.yaml

@@ -0,0 +1,54 @@
+# YOLOv5 🚀 by Ultralytics, AGPL-3.0 license
+
+# Parameters
+nc: 80  # number of classes
+depth_multiple: 1.0  # model depth multiple
+width_multiple: 1.0  # layer channel multiple
+anchors: 3  # AutoAnchor evolves 3 anchors per P output layer
+
+# YOLOv5 v6.0 backbone
+backbone:
+  # [from, number, module, args]
+  [[-1, 1, Conv, [64, 6, 2, 2]],  # 0-P1/2
+   [-1, 1, Conv, [128, 3, 2]],  # 1-P2/4
+   [-1, 3, C3, [128]],
+   [-1, 1, Conv, [256, 3, 2]],  # 3-P3/8
+   [-1, 6, C3, [256]],
+   [-1, 1, Conv, [512, 3, 2]],  # 5-P4/16
+   [-1, 9, C3, [512]],
+   [-1, 1, Conv, [1024, 3, 2]],  # 7-P5/32
+   [-1, 3, C3, [1024]],
+   [-1, 1, SPPF, [1024, 5]],  # 9
+  ]
+
+# YOLOv5 v6.0 head with (P2, P3, P4, P5) outputs
+head:
+  [[-1, 1, Conv, [512, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 6], 1, Concat, [1]],  # cat backbone P4
+   [-1, 3, C3, [512, False]],  # 13
+
+   [-1, 1, Conv, [256, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 4], 1, Concat, [1]],  # cat backbone P3
+   [-1, 3, C3, [256, False]],  # 17 (P3/8-small)
+
+   [-1, 1, Conv, [128, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 2], 1, Concat, [1]],  # cat backbone P2
+   [-1, 1, C3, [128, False]],  # 21 (P2/4-xsmall)
+
+   [-1, 1, Conv, [128, 3, 2]],
+   [[-1, 18], 1, Concat, [1]],  # cat head P3
+   [-1, 3, C3, [256, False]],  # 24 (P3/8-small)
+
+   [-1, 1, Conv, [256, 3, 2]],
+   [[-1, 14], 1, Concat, [1]],  # cat head P4
+   [-1, 3, C3, [512, False]],  # 27 (P4/16-medium)
+
+   [-1, 1, Conv, [512, 3, 2]],
+   [[-1, 10], 1, Concat, [1]],  # cat head P5
+   [-1, 3, C3, [1024, False]],  # 30 (P5/32-large)
+
+   [[21, 24, 27, 30], 1, Detect, [nc, anchors]],  # Detect(P2, P3, P4, P5)
+  ]

models/hub/yolov5-p34.yaml

@@ -0,0 +1,41 @@
+# YOLOv5 🚀 by Ultralytics, AGPL-3.0 license
+
+# Parameters
+nc: 80  # number of classes
+depth_multiple: 0.33  # model depth multiple
+width_multiple: 0.50  # layer channel multiple
+anchors: 3  # AutoAnchor evolves 3 anchors per P output layer
+
+# YOLOv5 v6.0 backbone
+backbone:
+  # [from, number, module, args]
+  [ [ -1, 1, Conv, [ 64, 6, 2, 2 ] ],  # 0-P1/2
+    [ -1, 1, Conv, [ 128, 3, 2 ] ],  # 1-P2/4
+    [ -1, 3, C3, [ 128 ] ],
+    [ -1, 1, Conv, [ 256, 3, 2 ] ],  # 3-P3/8
+    [ -1, 6, C3, [ 256 ] ],
+    [ -1, 1, Conv, [ 512, 3, 2 ] ],  # 5-P4/16
+    [ -1, 9, C3, [ 512 ] ],
+    [ -1, 1, Conv, [ 1024, 3, 2 ] ],  # 7-P5/32
+    [ -1, 3, C3, [ 1024 ] ],
+    [ -1, 1, SPPF, [ 1024, 5 ] ],  # 9
+  ]
+
+# YOLOv5 v6.0 head with (P3, P4) outputs
+head:
+  [ [ -1, 1, Conv, [ 512, 1, 1 ] ],
+    [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ],
+    [ [ -1, 6 ], 1, Concat, [ 1 ] ],  # cat backbone P4
+    [ -1, 3, C3, [ 512, False ] ],  # 13
+
+    [ -1, 1, Conv, [ 256, 1, 1 ] ],
+    [ -1, 1, nn.Upsample, [ None, 2, 'nearest' ] ],
+    [ [ -1, 4 ], 1, Concat, [ 1 ] ],  # cat backbone P3
+    [ -1, 3, C3, [ 256, False ] ],  # 17 (P3/8-small)
+
+    [ -1, 1, Conv, [ 256, 3, 2 ] ],
+    [ [ -1, 14 ], 1, Concat, [ 1 ] ],  # cat head P4
+    [ -1, 3, C3, [ 512, False ] ],  # 20 (P4/16-medium)
+
+    [ [ 17, 20 ], 1, Detect, [ nc, anchors ] ],  # Detect(P3, P4)
+  ]

models/hub/yolov5-p6.yaml

@@ -0,0 +1,56 @@
+# YOLOv5 🚀 by Ultralytics, AGPL-3.0 license
+
+# Parameters
+nc: 80  # number of classes
+depth_multiple: 1.0  # model depth multiple
+width_multiple: 1.0  # layer channel multiple
+anchors: 3  # AutoAnchor evolves 3 anchors per P output layer
+
+# YOLOv5 v6.0 backbone
+backbone:
+  # [from, number, module, args]
+  [[-1, 1, Conv, [64, 6, 2, 2]],  # 0-P1/2
+   [-1, 1, Conv, [128, 3, 2]],  # 1-P2/4
+   [-1, 3, C3, [128]],
+   [-1, 1, Conv, [256, 3, 2]],  # 3-P3/8
+   [-1, 6, C3, [256]],
+   [-1, 1, Conv, [512, 3, 2]],  # 5-P4/16
+   [-1, 9, C3, [512]],
+   [-1, 1, Conv, [768, 3, 2]],  # 7-P5/32
+   [-1, 3, C3, [768]],
+   [-1, 1, Conv, [1024, 3, 2]],  # 9-P6/64
+   [-1, 3, C3, [1024]],
+   [-1, 1, SPPF, [1024, 5]],  # 11
+  ]
+
+# YOLOv5 v6.0 head with (P3, P4, P5, P6) outputs
+head:
+  [[-1, 1, Conv, [768, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 8], 1, Concat, [1]],  # cat backbone P5
+   [-1, 3, C3, [768, False]],  # 15
+
+   [-1, 1, Conv, [512, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 6], 1, Concat, [1]],  # cat backbone P4
+   [-1, 3, C3, [512, False]],  # 19
+
+   [-1, 1, Conv, [256, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 4], 1, Concat, [1]],  # cat backbone P3
+   [-1, 3, C3, [256, False]],  # 23 (P3/8-small)
+
+   [-1, 1, Conv, [256, 3, 2]],
+   [[-1, 20], 1, Concat, [1]],  # cat head P4
+   [-1, 3, C3, [512, False]],  # 26 (P4/16-medium)
+
+   [-1, 1, Conv, [512, 3, 2]],
+   [[-1, 16], 1, Concat, [1]],  # cat head P5
+   [-1, 3, C3, [768, False]],  # 29 (P5/32-large)
+
+   [-1, 1, Conv, [768, 3, 2]],
+   [[-1, 12], 1, Concat, [1]],  # cat head P6
+   [-1, 3, C3, [1024, False]],  # 32 (P6/64-xlarge)
+
+   [[23, 26, 29, 32], 1, Detect, [nc, anchors]],  # Detect(P3, P4, P5, P6)
+  ]

models/hub/yolov5-p7.yaml

@@ -0,0 +1,67 @@
+# YOLOv5 🚀 by Ultralytics, AGPL-3.0 license
+
+# Parameters
+nc: 80  # number of classes
+depth_multiple: 1.0  # model depth multiple
+width_multiple: 1.0  # layer channel multiple
+anchors: 3  # AutoAnchor evolves 3 anchors per P output layer
+
+# YOLOv5 v6.0 backbone
+backbone:
+  # [from, number, module, args]
+  [[-1, 1, Conv, [64, 6, 2, 2]],  # 0-P1/2
+   [-1, 1, Conv, [128, 3, 2]],  # 1-P2/4
+   [-1, 3, C3, [128]],
+   [-1, 1, Conv, [256, 3, 2]],  # 3-P3/8
+   [-1, 6, C3, [256]],
+   [-1, 1, Conv, [512, 3, 2]],  # 5-P4/16
+   [-1, 9, C3, [512]],
+   [-1, 1, Conv, [768, 3, 2]],  # 7-P5/32
+   [-1, 3, C3, [768]],
+   [-1, 1, Conv, [1024, 3, 2]],  # 9-P6/64
+   [-1, 3, C3, [1024]],
+   [-1, 1, Conv, [1280, 3, 2]],  # 11-P7/128
+   [-1, 3, C3, [1280]],
+   [-1, 1, SPPF, [1280, 5]],  # 13
+  ]
+
+# YOLOv5 v6.0 head with (P3, P4, P5, P6, P7) outputs
+head:
+  [[-1, 1, Conv, [1024, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 10], 1, Concat, [1]],  # cat backbone P6
+   [-1, 3, C3, [1024, False]],  # 17
+
+   [-1, 1, Conv, [768, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 8], 1, Concat, [1]],  # cat backbone P5
+   [-1, 3, C3, [768, False]],  # 21
+
+   [-1, 1, Conv, [512, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 6], 1, Concat, [1]],  # cat backbone P4
+   [-1, 3, C3, [512, False]],  # 25
+
+   [-1, 1, Conv, [256, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 4], 1, Concat, [1]],  # cat backbone P3
+   [-1, 3, C3, [256, False]],  # 29 (P3/8-small)
+
+   [-1, 1, Conv, [256, 3, 2]],
+   [[-1, 26], 1, Concat, [1]],  # cat head P4
+   [-1, 3, C3, [512, False]],  # 32 (P4/16-medium)
+
+   [-1, 1, Conv, [512, 3, 2]],
+   [[-1, 22], 1, Concat, [1]],  # cat head P5
+   [-1, 3, C3, [768, False]],  # 35 (P5/32-large)
+
+   [-1, 1, Conv, [768, 3, 2]],
+   [[-1, 18], 1, Concat, [1]],  # cat head P6
+   [-1, 3, C3, [1024, False]],  # 38 (P6/64-xlarge)
+
+   [-1, 1, Conv, [1024, 3, 2]],
+   [[-1, 14], 1, Concat, [1]],  # cat head P7
+   [-1, 3, C3, [1280, False]],  # 41 (P7/128-xxlarge)
+
+   [[29, 32, 35, 38, 41], 1, Detect, [nc, anchors]],  # Detect(P3, P4, P5, P6, P7)
+  ]

models/hub/yolov5-panet.yaml

@@ -0,0 +1,48 @@
+# YOLOv5 🚀 by Ultralytics, AGPL-3.0 license
+
+# Parameters
+nc: 80  # number of classes
+depth_multiple: 1.0  # model depth multiple
+width_multiple: 1.0  # layer channel multiple
+anchors:
+  - [10,13, 16,30, 33,23]  # P3/8
+  - [30,61, 62,45, 59,119]  # P4/16
+  - [116,90, 156,198, 373,326]  # P5/32
+
+# YOLOv5 v6.0 backbone
+backbone:
+  # [from, number, module, args]
+  [[-1, 1, Conv, [64, 6, 2, 2]],  # 0-P1/2
+   [-1, 1, Conv, [128, 3, 2]],  # 1-P2/4
+   [-1, 3, C3, [128]],
+   [-1, 1, Conv, [256, 3, 2]],  # 3-P3/8
+   [-1, 6, C3, [256]],
+   [-1, 1, Conv, [512, 3, 2]],  # 5-P4/16
+   [-1, 9, C3, [512]],
+   [-1, 1, Conv, [1024, 3, 2]],  # 7-P5/32
+   [-1, 3, C3, [1024]],
+   [-1, 1, SPPF, [1024, 5]],  # 9
+  ]
+
+# YOLOv5 v6.0 PANet head
+head:
+  [[-1, 1, Conv, [512, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 6], 1, Concat, [1]],  # cat backbone P4
+   [-1, 3, C3, [512, False]],  # 13
+
+   [-1, 1, Conv, [256, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 4], 1, Concat, [1]],  # cat backbone P3
+   [-1, 3, C3, [256, False]],  # 17 (P3/8-small)
+
+   [-1, 1, Conv, [256, 3, 2]],
+   [[-1, 14], 1, Concat, [1]],  # cat head P4
+   [-1, 3, C3, [512, False]],  # 20 (P4/16-medium)
+
+   [-1, 1, Conv, [512, 3, 2]],
+   [[-1, 10], 1, Concat, [1]],  # cat head P5
+   [-1, 3, C3, [1024, False]],  # 23 (P5/32-large)
+
+   [[17, 20, 23], 1, Detect, [nc, anchors]],  # Detect(P3, P4, P5)
+  ]

models/hub/yolov5l6.yaml

@@ -0,0 +1,60 @@
+# YOLOv5 🚀 by Ultralytics, AGPL-3.0 license
+
+# Parameters
+nc: 80  # number of classes
+depth_multiple: 1.0  # model depth multiple
+width_multiple: 1.0  # layer channel multiple
+anchors:
+  - [19,27,  44,40,  38,94]  # P3/8
+  - [96,68,  86,152,  180,137]  # P4/16
+  - [140,301,  303,264,  238,542]  # P5/32
+  - [436,615,  739,380,  925,792]  # P6/64
+
+# YOLOv5 v6.0 backbone
+backbone:
+  # [from, number, module, args]
+  [[-1, 1, Conv, [64, 6, 2, 2]],  # 0-P1/2
+   [-1, 1, Conv, [128, 3, 2]],  # 1-P2/4
+   [-1, 3, C3, [128]],
+   [-1, 1, Conv, [256, 3, 2]],  # 3-P3/8
+   [-1, 6, C3, [256]],
+   [-1, 1, Conv, [512, 3, 2]],  # 5-P4/16
+   [-1, 9, C3, [512]],
+   [-1, 1, Conv, [768, 3, 2]],  # 7-P5/32
+   [-1, 3, C3, [768]],
+   [-1, 1, Conv, [1024, 3, 2]],  # 9-P6/64
+   [-1, 3, C3, [1024]],
+   [-1, 1, SPPF, [1024, 5]],  # 11
+  ]
+
+# YOLOv5 v6.0 head
+head:
+  [[-1, 1, Conv, [768, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 8], 1, Concat, [1]],  # cat backbone P5
+   [-1, 3, C3, [768, False]],  # 15
+
+   [-1, 1, Conv, [512, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 6], 1, Concat, [1]],  # cat backbone P4
+   [-1, 3, C3, [512, False]],  # 19
+
+   [-1, 1, Conv, [256, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 4], 1, Concat, [1]],  # cat backbone P3
+   [-1, 3, C3, [256, False]],  # 23 (P3/8-small)
+
+   [-1, 1, Conv, [256, 3, 2]],
+   [[-1, 20], 1, Concat, [1]],  # cat head P4
+   [-1, 3, C3, [512, False]],  # 26 (P4/16-medium)
+
+   [-1, 1, Conv, [512, 3, 2]],
+   [[-1, 16], 1, Concat, [1]],  # cat head P5
+   [-1, 3, C3, [768, False]],  # 29 (P5/32-large)
+
+   [-1, 1, Conv, [768, 3, 2]],
+   [[-1, 12], 1, Concat, [1]],  # cat head P6
+   [-1, 3, C3, [1024, False]],  # 32 (P6/64-xlarge)
+
+   [[23, 26, 29, 32], 1, Detect, [nc, anchors]],  # Detect(P3, P4, P5, P6)
+  ]

models/hub/yolov5m6.yaml

@@ -0,0 +1,60 @@
+# YOLOv5 🚀 by Ultralytics, AGPL-3.0 license
+
+# Parameters
+nc: 80  # number of classes
+depth_multiple: 0.67  # model depth multiple
+width_multiple: 0.75  # layer channel multiple
+anchors:
+  - [19,27,  44,40,  38,94]  # P3/8
+  - [96,68,  86,152,  180,137]  # P4/16
+  - [140,301,  303,264,  238,542]  # P5/32
+  - [436,615,  739,380,  925,792]  # P6/64
+
+# YOLOv5 v6.0 backbone
+backbone:
+  # [from, number, module, args]
+  [[-1, 1, Conv, [64, 6, 2, 2]],  # 0-P1/2
+   [-1, 1, Conv, [128, 3, 2]],  # 1-P2/4
+   [-1, 3, C3, [128]],
+   [-1, 1, Conv, [256, 3, 2]],  # 3-P3/8
+   [-1, 6, C3, [256]],
+   [-1, 1, Conv, [512, 3, 2]],  # 5-P4/16
+   [-1, 9, C3, [512]],
+   [-1, 1, Conv, [768, 3, 2]],  # 7-P5/32
+   [-1, 3, C3, [768]],
+   [-1, 1, Conv, [1024, 3, 2]],  # 9-P6/64
+   [-1, 3, C3, [1024]],
+   [-1, 1, SPPF, [1024, 5]],  # 11
+  ]
+
+# YOLOv5 v6.0 head
+head:
+  [[-1, 1, Conv, [768, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 8], 1, Concat, [1]],  # cat backbone P5
+   [-1, 3, C3, [768, False]],  # 15
+
+   [-1, 1, Conv, [512, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 6], 1, Concat, [1]],  # cat backbone P4
+   [-1, 3, C3, [512, False]],  # 19
+
+   [-1, 1, Conv, [256, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 4], 1, Concat, [1]],  # cat backbone P3
+   [-1, 3, C3, [256, False]],  # 23 (P3/8-small)
+
+   [-1, 1, Conv, [256, 3, 2]],
+   [[-1, 20], 1, Concat, [1]],  # cat head P4
+   [-1, 3, C3, [512, False]],  # 26 (P4/16-medium)
+
+   [-1, 1, Conv, [512, 3, 2]],
+   [[-1, 16], 1, Concat, [1]],  # cat head P5
+   [-1, 3, C3, [768, False]],  # 29 (P5/32-large)
+
+   [-1, 1, Conv, [768, 3, 2]],
+   [[-1, 12], 1, Concat, [1]],  # cat head P6
+   [-1, 3, C3, [1024, False]],  # 32 (P6/64-xlarge)
+
+   [[23, 26, 29, 32], 1, Detect, [nc, anchors]],  # Detect(P3, P4, P5, P6)
+  ]

models/hub/yolov5n6.yaml

@@ -0,0 +1,60 @@
+# YOLOv5 🚀 by Ultralytics, AGPL-3.0 license
+
+# Parameters
+nc: 80  # number of classes
+depth_multiple: 0.33  # model depth multiple
+width_multiple: 0.25  # layer channel multiple
+anchors:
+  - [19,27,  44,40,  38,94]  # P3/8
+  - [96,68,  86,152,  180,137]  # P4/16
+  - [140,301,  303,264,  238,542]  # P5/32
+  - [436,615,  739,380,  925,792]  # P6/64
+
+# YOLOv5 v6.0 backbone
+backbone:
+  # [from, number, module, args]
+  [[-1, 1, Conv, [64, 6, 2, 2]],  # 0-P1/2
+   [-1, 1, Conv, [128, 3, 2]],  # 1-P2/4
+   [-1, 3, C3, [128]],
+   [-1, 1, Conv, [256, 3, 2]],  # 3-P3/8
+   [-1, 6, C3, [256]],
+   [-1, 1, Conv, [512, 3, 2]],  # 5-P4/16
+   [-1, 9, C3, [512]],
+   [-1, 1, Conv, [768, 3, 2]],  # 7-P5/32
+   [-1, 3, C3, [768]],
+   [-1, 1, Conv, [1024, 3, 2]],  # 9-P6/64
+   [-1, 3, C3, [1024]],
+   [-1, 1, SPPF, [1024, 5]],  # 11
+  ]
+
+# YOLOv5 v6.0 head
+head:
+  [[-1, 1, Conv, [768, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 8], 1, Concat, [1]],  # cat backbone P5
+   [-1, 3, C3, [768, False]],  # 15
+
+   [-1, 1, Conv, [512, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 6], 1, Concat, [1]],  # cat backbone P4
+   [-1, 3, C3, [512, False]],  # 19
+
+   [-1, 1, Conv, [256, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 4], 1, Concat, [1]],  # cat backbone P3
+   [-1, 3, C3, [256, False]],  # 23 (P3/8-small)
+
+   [-1, 1, Conv, [256, 3, 2]],
+   [[-1, 20], 1, Concat, [1]],  # cat head P4
+   [-1, 3, C3, [512, False]],  # 26 (P4/16-medium)
+
+   [-1, 1, Conv, [512, 3, 2]],
+   [[-1, 16], 1, Concat, [1]],  # cat head P5
+   [-1, 3, C3, [768, False]],  # 29 (P5/32-large)
+
+   [-1, 1, Conv, [768, 3, 2]],
+   [[-1, 12], 1, Concat, [1]],  # cat head P6
+   [-1, 3, C3, [1024, False]],  # 32 (P6/64-xlarge)
+
+   [[23, 26, 29, 32], 1, Detect, [nc, anchors]],  # Detect(P3, P4, P5, P6)
+  ]

models/hub/yolov5s-LeakyReLU.yaml

@@ -0,0 +1,49 @@
+# YOLOv5 🚀 by Ultralytics, AGPL-3.0 license
+
+# Parameters
+nc: 80  # number of classes
+activation: nn.LeakyReLU(0.1)  # <----- Conv() activation used throughout entire YOLOv5 model
+depth_multiple: 0.33  # model depth multiple
+width_multiple: 0.50  # layer channel multiple
+anchors:
+  - [10,13, 16,30, 33,23]  # P3/8
+  - [30,61, 62,45, 59,119]  # P4/16
+  - [116,90, 156,198, 373,326]  # P5/32
+
+# YOLOv5 v6.0 backbone
+backbone:
+  # [from, number, module, args]
+  [[-1, 1, Conv, [64, 6, 2, 2]],  # 0-P1/2
+   [-1, 1, Conv, [128, 3, 2]],  # 1-P2/4
+   [-1, 3, C3, [128]],
+   [-1, 1, Conv, [256, 3, 2]],  # 3-P3/8
+   [-1, 6, C3, [256]],
+   [-1, 1, Conv, [512, 3, 2]],  # 5-P4/16
+   [-1, 9, C3, [512]],
+   [-1, 1, Conv, [1024, 3, 2]],  # 7-P5/32
+   [-1, 3, C3, [1024]],
+   [-1, 1, SPPF, [1024, 5]],  # 9
+  ]
+
+# YOLOv5 v6.0 head
+head:
+  [[-1, 1, Conv, [512, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 6], 1, Concat, [1]],  # cat backbone P4
+   [-1, 3, C3, [512, False]],  # 13
+
+   [-1, 1, Conv, [256, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 4], 1, Concat, [1]],  # cat backbone P3
+   [-1, 3, C3, [256, False]],  # 17 (P3/8-small)
+
+   [-1, 1, Conv, [256, 3, 2]],
+   [[-1, 14], 1, Concat, [1]],  # cat head P4
+   [-1, 3, C3, [512, False]],  # 20 (P4/16-medium)
+
+   [-1, 1, Conv, [512, 3, 2]],
+   [[-1, 10], 1, Concat, [1]],  # cat head P5
+   [-1, 3, C3, [1024, False]],  # 23 (P5/32-large)
+
+   [[17, 20, 23], 1, Detect, [nc, anchors]],  # Detect(P3, P4, P5)
+  ]

models/hub/yolov5s-ghost.yaml

@@ -0,0 +1,48 @@
+# YOLOv5 🚀 by Ultralytics, AGPL-3.0 license
+
+# Parameters
+nc: 80  # number of classes
+depth_multiple: 0.33  # model depth multiple
+width_multiple: 0.50  # layer channel multiple
+anchors:
+  - [10,13, 16,30, 33,23]  # P3/8
+  - [30,61, 62,45, 59,119]  # P4/16
+  - [116,90, 156,198, 373,326]  # P5/32
+
+# YOLOv5 v6.0 backbone
+backbone:
+  # [from, number, module, args]
+  [[-1, 1, Conv, [64, 6, 2, 2]],  # 0-P1/2
+   [-1, 1, GhostConv, [128, 3, 2]],  # 1-P2/4
+   [-1, 3, C3Ghost, [128]],
+   [-1, 1, GhostConv, [256, 3, 2]],  # 3-P3/8
+   [-1, 6, C3Ghost, [256]],
+   [-1, 1, GhostConv, [512, 3, 2]],  # 5-P4/16
+   [-1, 9, C3Ghost, [512]],
+   [-1, 1, GhostConv, [1024, 3, 2]],  # 7-P5/32
+   [-1, 3, C3Ghost, [1024]],
+   [-1, 1, SPPF, [1024, 5]],  # 9
+  ]
+
+# YOLOv5 v6.0 head
+head:
+  [[-1, 1, GhostConv, [512, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 6], 1, Concat, [1]],  # cat backbone P4
+   [-1, 3, C3Ghost, [512, False]],  # 13
+
+   [-1, 1, GhostConv, [256, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 4], 1, Concat, [1]],  # cat backbone P3
+   [-1, 3, C3Ghost, [256, False]],  # 17 (P3/8-small)
+
+   [-1, 1, GhostConv, [256, 3, 2]],
+   [[-1, 14], 1, Concat, [1]],  # cat head P4
+   [-1, 3, C3Ghost, [512, False]],  # 20 (P4/16-medium)
+
+   [-1, 1, GhostConv, [512, 3, 2]],
+   [[-1, 10], 1, Concat, [1]],  # cat head P5
+   [-1, 3, C3Ghost, [1024, False]],  # 23 (P5/32-large)
+
+   [[17, 20, 23], 1, Detect, [nc, anchors]],  # Detect(P3, P4, P5)
+  ]

models/hub/yolov5s-transformer.yaml

@@ -0,0 +1,48 @@
+# YOLOv5 🚀 by Ultralytics, AGPL-3.0 license
+
+# Parameters
+nc: 80  # number of classes
+depth_multiple: 0.33  # model depth multiple
+width_multiple: 0.50  # layer channel multiple
+anchors:
+  - [10,13, 16,30, 33,23]  # P3/8
+  - [30,61, 62,45, 59,119]  # P4/16
+  - [116,90, 156,198, 373,326]  # P5/32
+
+# YOLOv5 v6.0 backbone
+backbone:
+  # [from, number, module, args]
+  [[-1, 1, Conv, [64, 6, 2, 2]],  # 0-P1/2
+   [-1, 1, Conv, [128, 3, 2]],  # 1-P2/4
+   [-1, 3, C3, [128]],
+   [-1, 1, Conv, [256, 3, 2]],  # 3-P3/8
+   [-1, 6, C3, [256]],
+   [-1, 1, Conv, [512, 3, 2]],  # 5-P4/16
+   [-1, 9, C3, [512]],
+   [-1, 1, Conv, [1024, 3, 2]],  # 7-P5/32
+   [-1, 3, C3TR, [1024]],  # 9 <--- C3TR() Transformer module
+   [-1, 1, SPPF, [1024, 5]],  # 9
+  ]
+
+# YOLOv5 v6.0 head
+head:
+  [[-1, 1, Conv, [512, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 6], 1, Concat, [1]],  # cat backbone P4
+   [-1, 3, C3, [512, False]],  # 13
+
+   [-1, 1, Conv, [256, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 4], 1, Concat, [1]],  # cat backbone P3
+   [-1, 3, C3, [256, False]],  # 17 (P3/8-small)
+
+   [-1, 1, Conv, [256, 3, 2]],
+   [[-1, 14], 1, Concat, [1]],  # cat head P4
+   [-1, 3, C3, [512, False]],  # 20 (P4/16-medium)
+
+   [-1, 1, Conv, [512, 3, 2]],
+   [[-1, 10], 1, Concat, [1]],  # cat head P5
+   [-1, 3, C3, [1024, False]],  # 23 (P5/32-large)
+
+   [[17, 20, 23], 1, Detect, [nc, anchors]],  # Detect(P3, P4, P5)
+  ]

models/hub/yolov5s6.yaml

@@ -0,0 +1,60 @@
+# YOLOv5 🚀 by Ultralytics, AGPL-3.0 license
+
+# Parameters
+nc: 80  # number of classes
+depth_multiple: 0.33  # model depth multiple
+width_multiple: 0.50  # layer channel multiple
+anchors:
+  - [19,27,  44,40,  38,94]  # P3/8
+  - [96,68,  86,152,  180,137]  # P4/16
+  - [140,301,  303,264,  238,542]  # P5/32
+  - [436,615,  739,380,  925,792]  # P6/64
+
+# YOLOv5 v6.0 backbone
+backbone:
+  # [from, number, module, args]
+  [[-1, 1, Conv, [64, 6, 2, 2]],  # 0-P1/2
+   [-1, 1, Conv, [128, 3, 2]],  # 1-P2/4
+   [-1, 3, C3, [128]],
+   [-1, 1, Conv, [256, 3, 2]],  # 3-P3/8
+   [-1, 6, C3, [256]],
+   [-1, 1, Conv, [512, 3, 2]],  # 5-P4/16
+   [-1, 9, C3, [512]],
+   [-1, 1, Conv, [768, 3, 2]],  # 7-P5/32
+   [-1, 3, C3, [768]],
+   [-1, 1, Conv, [1024, 3, 2]],  # 9-P6/64
+   [-1, 3, C3, [1024]],
+   [-1, 1, SPPF, [1024, 5]],  # 11
+  ]
+
+# YOLOv5 v6.0 head
+head:
+  [[-1, 1, Conv, [768, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 8], 1, Concat, [1]],  # cat backbone P5
+   [-1, 3, C3, [768, False]],  # 15
+
+   [-1, 1, Conv, [512, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 6], 1, Concat, [1]],  # cat backbone P4
+   [-1, 3, C3, [512, False]],  # 19
+
+   [-1, 1, Conv, [256, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 4], 1, Concat, [1]],  # cat backbone P3
+   [-1, 3, C3, [256, False]],  # 23 (P3/8-small)
+
+   [-1, 1, Conv, [256, 3, 2]],
+   [[-1, 20], 1, Concat, [1]],  # cat head P4
+   [-1, 3, C3, [512, False]],  # 26 (P4/16-medium)
+
+   [-1, 1, Conv, [512, 3, 2]],
+   [[-1, 16], 1, Concat, [1]],  # cat head P5
+   [-1, 3, C3, [768, False]],  # 29 (P5/32-large)
+
+   [-1, 1, Conv, [768, 3, 2]],
+   [[-1, 12], 1, Concat, [1]],  # cat head P6
+   [-1, 3, C3, [1024, False]],  # 32 (P6/64-xlarge)
+
+   [[23, 26, 29, 32], 1, Detect, [nc, anchors]],  # Detect(P3, P4, P5, P6)
+  ]

models/hub/yolov5x6.yaml

@@ -0,0 +1,60 @@
+# YOLOv5 🚀 by Ultralytics, AGPL-3.0 license
+
+# Parameters
+nc: 80  # number of classes
+depth_multiple: 1.33  # model depth multiple
+width_multiple: 1.25  # layer channel multiple
+anchors:
+  - [19,27,  44,40,  38,94]  # P3/8
+  - [96,68,  86,152,  180,137]  # P4/16
+  - [140,301,  303,264,  238,542]  # P5/32
+  - [436,615,  739,380,  925,792]  # P6/64
+
+# YOLOv5 v6.0 backbone
+backbone:
+  # [from, number, module, args]
+  [[-1, 1, Conv, [64, 6, 2, 2]],  # 0-P1/2
+   [-1, 1, Conv, [128, 3, 2]],  # 1-P2/4
+   [-1, 3, C3, [128]],
+   [-1, 1, Conv, [256, 3, 2]],  # 3-P3/8
+   [-1, 6, C3, [256]],
+   [-1, 1, Conv, [512, 3, 2]],  # 5-P4/16
+   [-1, 9, C3, [512]],
+   [-1, 1, Conv, [768, 3, 2]],  # 7-P5/32
+   [-1, 3, C3, [768]],
+   [-1, 1, Conv, [1024, 3, 2]],  # 9-P6/64
+   [-1, 3, C3, [1024]],
+   [-1, 1, SPPF, [1024, 5]],  # 11
+  ]
+
+# YOLOv5 v6.0 head
+head:
+  [[-1, 1, Conv, [768, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 8], 1, Concat, [1]],  # cat backbone P5
+   [-1, 3, C3, [768, False]],  # 15
+
+   [-1, 1, Conv, [512, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 6], 1, Concat, [1]],  # cat backbone P4
+   [-1, 3, C3, [512, False]],  # 19
+
+   [-1, 1, Conv, [256, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 4], 1, Concat, [1]],  # cat backbone P3
+   [-1, 3, C3, [256, False]],  # 23 (P3/8-small)
+
+   [-1, 1, Conv, [256, 3, 2]],
+   [[-1, 20], 1, Concat, [1]],  # cat head P4
+   [-1, 3, C3, [512, False]],  # 26 (P4/16-medium)
+
+   [-1, 1, Conv, [512, 3, 2]],
+   [[-1, 16], 1, Concat, [1]],  # cat head P5
+   [-1, 3, C3, [768, False]],  # 29 (P5/32-large)
+
+   [-1, 1, Conv, [768, 3, 2]],
+   [[-1, 12], 1, Concat, [1]],  # cat head P6
+   [-1, 3, C3, [1024, False]],  # 32 (P6/64-xlarge)
+
+   [[23, 26, 29, 32], 1, Detect, [nc, anchors]],  # Detect(P3, P4, P5, P6)
+  ]

models/segment/yolov5l-seg.yaml

@@ -0,0 +1,48 @@
+# YOLOv5 🚀 by Ultralytics, AGPL-3.0 license
+
+# Parameters
+nc: 80  # number of classes
+depth_multiple: 1.0  # model depth multiple
+width_multiple: 1.0  # layer channel multiple
+anchors:
+  - [10,13, 16,30, 33,23]  # P3/8
+  - [30,61, 62,45, 59,119]  # P4/16
+  - [116,90, 156,198, 373,326]  # P5/32
+
+# YOLOv5 v6.0 backbone
+backbone:
+  # [from, number, module, args]
+  [[-1, 1, Conv, [64, 6, 2, 2]],  # 0-P1/2
+   [-1, 1, Conv, [128, 3, 2]],  # 1-P2/4
+   [-1, 3, C3, [128]],
+   [-1, 1, Conv, [256, 3, 2]],  # 3-P3/8
+   [-1, 6, C3, [256]],
+   [-1, 1, Conv, [512, 3, 2]],  # 5-P4/16
+   [-1, 9, C3, [512]],
+   [-1, 1, Conv, [1024, 3, 2]],  # 7-P5/32
+   [-1, 3, C3, [1024]],
+   [-1, 1, SPPF, [1024, 5]],  # 9
+  ]
+
+# YOLOv5 v6.0 head
+head:
+  [[-1, 1, Conv, [512, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 6], 1, Concat, [1]],  # cat backbone P4
+   [-1, 3, C3, [512, False]],  # 13
+
+   [-1, 1, Conv, [256, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 4], 1, Concat, [1]],  # cat backbone P3
+   [-1, 3, C3, [256, False]],  # 17 (P3/8-small)
+
+   [-1, 1, Conv, [256, 3, 2]],
+   [[-1, 14], 1, Concat, [1]],  # cat head P4
+   [-1, 3, C3, [512, False]],  # 20 (P4/16-medium)
+
+   [-1, 1, Conv, [512, 3, 2]],
+   [[-1, 10], 1, Concat, [1]],  # cat head P5
+   [-1, 3, C3, [1024, False]],  # 23 (P5/32-large)
+
+   [[17, 20, 23], 1, Segment, [nc, anchors, 32, 256]],  # Detect(P3, P4, P5)
+  ]

models/segment/yolov5m-seg.yaml

@@ -0,0 +1,48 @@
+# YOLOv5 🚀 by Ultralytics, AGPL-3.0 license
+
+# Parameters
+nc: 80  # number of classes
+depth_multiple: 0.67  # model depth multiple
+width_multiple: 0.75  # layer channel multiple
+anchors:
+  - [10,13, 16,30, 33,23]  # P3/8
+  - [30,61, 62,45, 59,119]  # P4/16
+  - [116,90, 156,198, 373,326]  # P5/32
+
+# YOLOv5 v6.0 backbone
+backbone:
+  # [from, number, module, args]
+  [[-1, 1, Conv, [64, 6, 2, 2]],  # 0-P1/2
+   [-1, 1, Conv, [128, 3, 2]],  # 1-P2/4
+   [-1, 3, C3, [128]],
+   [-1, 1, Conv, [256, 3, 2]],  # 3-P3/8
+   [-1, 6, C3, [256]],
+   [-1, 1, Conv, [512, 3, 2]],  # 5-P4/16
+   [-1, 9, C3, [512]],
+   [-1, 1, Conv, [1024, 3, 2]],  # 7-P5/32
+   [-1, 3, C3, [1024]],
+   [-1, 1, SPPF, [1024, 5]],  # 9
+  ]
+
+# YOLOv5 v6.0 head
+head:
+  [[-1, 1, Conv, [512, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 6], 1, Concat, [1]],  # cat backbone P4
+   [-1, 3, C3, [512, False]],  # 13
+
+   [-1, 1, Conv, [256, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 4], 1, Concat, [1]],  # cat backbone P3
+   [-1, 3, C3, [256, False]],  # 17 (P3/8-small)
+
+   [-1, 1, Conv, [256, 3, 2]],
+   [[-1, 14], 1, Concat, [1]],  # cat head P4
+   [-1, 3, C3, [512, False]],  # 20 (P4/16-medium)
+
+   [-1, 1, Conv, [512, 3, 2]],
+   [[-1, 10], 1, Concat, [1]],  # cat head P5
+   [-1, 3, C3, [1024, False]],  # 23 (P5/32-large)
+
+   [[17, 20, 23], 1, Segment, [nc, anchors, 32, 256]],  # Detect(P3, P4, P5)
+  ]

models/segment/yolov5n-seg.yaml

@@ -0,0 +1,48 @@
+# YOLOv5 🚀 by Ultralytics, AGPL-3.0 license
+
+# Parameters
+nc: 80  # number of classes
+depth_multiple: 0.33  # model depth multiple
+width_multiple: 0.25  # layer channel multiple
+anchors:
+  - [10,13, 16,30, 33,23]  # P3/8
+  - [30,61, 62,45, 59,119]  # P4/16
+  - [116,90, 156,198, 373,326]  # P5/32
+
+# YOLOv5 v6.0 backbone
+backbone:
+  # [from, number, module, args]
+  [[-1, 1, Conv, [64, 6, 2, 2]],  # 0-P1/2
+   [-1, 1, Conv, [128, 3, 2]],  # 1-P2/4
+   [-1, 3, C3, [128]],
+   [-1, 1, Conv, [256, 3, 2]],  # 3-P3/8
+   [-1, 6, C3, [256]],
+   [-1, 1, Conv, [512, 3, 2]],  # 5-P4/16
+   [-1, 9, C3, [512]],
+   [-1, 1, Conv, [1024, 3, 2]],  # 7-P5/32
+   [-1, 3, C3, [1024]],
+   [-1, 1, SPPF, [1024, 5]],  # 9
+  ]
+
+# YOLOv5 v6.0 head
+head:
+  [[-1, 1, Conv, [512, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 6], 1, Concat, [1]],  # cat backbone P4
+   [-1, 3, C3, [512, False]],  # 13
+
+   [-1, 1, Conv, [256, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 4], 1, Concat, [1]],  # cat backbone P3
+   [-1, 3, C3, [256, False]],  # 17 (P3/8-small)
+
+   [-1, 1, Conv, [256, 3, 2]],
+   [[-1, 14], 1, Concat, [1]],  # cat head P4
+   [-1, 3, C3, [512, False]],  # 20 (P4/16-medium)
+
+   [-1, 1, Conv, [512, 3, 2]],
+   [[-1, 10], 1, Concat, [1]],  # cat head P5
+   [-1, 3, C3, [1024, False]],  # 23 (P5/32-large)
+
+   [[17, 20, 23], 1, Segment, [nc, anchors, 32, 256]],  # Detect(P3, P4, P5)
+  ]

models/segment/yolov5s-seg.yaml

@@ -0,0 +1,48 @@
+# YOLOv5 🚀 by Ultralytics, AGPL-3.0 license
+
+# Parameters
+nc: 80  # number of classes
+depth_multiple: 0.33  # model depth multiple
+width_multiple: 0.5  # layer channel multiple
+anchors:
+  - [10,13, 16,30, 33,23]  # P3/8
+  - [30,61, 62,45, 59,119]  # P4/16
+  - [116,90, 156,198, 373,326]  # P5/32
+
+# YOLOv5 v6.0 backbone
+backbone:
+  # [from, number, module, args]
+  [[-1, 1, Conv, [64, 6, 2, 2]],  # 0-P1/2
+   [-1, 1, Conv, [128, 3, 2]],  # 1-P2/4
+   [-1, 3, C3, [128]],
+   [-1, 1, Conv, [256, 3, 2]],  # 3-P3/8
+   [-1, 6, C3, [256]],
+   [-1, 1, Conv, [512, 3, 2]],  # 5-P4/16
+   [-1, 9, C3, [512]],
+   [-1, 1, Conv, [1024, 3, 2]],  # 7-P5/32
+   [-1, 3, C3, [1024]],
+   [-1, 1, SPPF, [1024, 5]],  # 9
+  ]
+
+# YOLOv5 v6.0 head
+head:
+  [[-1, 1, Conv, [512, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 6], 1, Concat, [1]],  # cat backbone P4
+   [-1, 3, C3, [512, False]],  # 13
+
+   [-1, 1, Conv, [256, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 4], 1, Concat, [1]],  # cat backbone P3
+   [-1, 3, C3, [256, False]],  # 17 (P3/8-small)
+
+   [-1, 1, Conv, [256, 3, 2]],
+   [[-1, 14], 1, Concat, [1]],  # cat head P4
+   [-1, 3, C3, [512, False]],  # 20 (P4/16-medium)
+
+   [-1, 1, Conv, [512, 3, 2]],
+   [[-1, 10], 1, Concat, [1]],  # cat head P5
+   [-1, 3, C3, [1024, False]],  # 23 (P5/32-large)
+
+   [[17, 20, 23], 1, Segment, [nc, anchors, 32, 256]],  # Detect(P3, P4, P5)
+  ]

models/segment/yolov5x-seg.yaml

@@ -0,0 +1,48 @@
+# YOLOv5 🚀 by Ultralytics, AGPL-3.0 license
+
+# Parameters
+nc: 80  # number of classes
+depth_multiple: 1.33  # model depth multiple
+width_multiple: 1.25  # layer channel multiple
+anchors:
+  - [10,13, 16,30, 33,23]  # P3/8
+  - [30,61, 62,45, 59,119]  # P4/16
+  - [116,90, 156,198, 373,326]  # P5/32
+
+# YOLOv5 v6.0 backbone
+backbone:
+  # [from, number, module, args]
+  [[-1, 1, Conv, [64, 6, 2, 2]],  # 0-P1/2
+   [-1, 1, Conv, [128, 3, 2]],  # 1-P2/4
+   [-1, 3, C3, [128]],
+   [-1, 1, Conv, [256, 3, 2]],  # 3-P3/8
+   [-1, 6, C3, [256]],
+   [-1, 1, Conv, [512, 3, 2]],  # 5-P4/16
+   [-1, 9, C3, [512]],
+   [-1, 1, Conv, [1024, 3, 2]],  # 7-P5/32
+   [-1, 3, C3, [1024]],
+   [-1, 1, SPPF, [1024, 5]],  # 9
+  ]
+
+# YOLOv5 v6.0 head
+head:
+  [[-1, 1, Conv, [512, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 6], 1, Concat, [1]],  # cat backbone P4
+   [-1, 3, C3, [512, False]],  # 13
+
+   [-1, 1, Conv, [256, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 4], 1, Concat, [1]],  # cat backbone P3
+   [-1, 3, C3, [256, False]],  # 17 (P3/8-small)
+
+   [-1, 1, Conv, [256, 3, 2]],
+   [[-1, 14], 1, Concat, [1]],  # cat head P4
+   [-1, 3, C3, [512, False]],  # 20 (P4/16-medium)
+
+   [-1, 1, Conv, [512, 3, 2]],
+   [[-1, 10], 1, Concat, [1]],  # cat head P5
+   [-1, 3, C3, [1024, False]],  # 23 (P5/32-large)
+
+   [[17, 20, 23], 1, Segment, [nc, anchors, 32, 256]],  # Detect(P3, P4, P5)
+  ]

models/tf.py

@@ -0,0 +1,608 @@
+# YOLOv5 🚀 by Ultralytics, AGPL-3.0 license
+"""
+TensorFlow, Keras and TFLite versions of YOLOv5
+Authored by https://github.com/zldrobit in PR https://github.com/ultralytics/yolov5/pull/1127
+
+Usage:
+    $ python models/tf.py --weights yolov5s.pt
+
+Export:
+    $ python export.py --weights yolov5s.pt --include saved_model pb tflite tfjs
+"""
+
+import argparse
+import sys
+from copy import deepcopy
+from pathlib import Path
+
+FILE = Path(__file__).resolve()
+ROOT = FILE.parents[1]  # YOLOv5 root directory
+if str(ROOT) not in sys.path:
+    sys.path.append(str(ROOT))  # add ROOT to PATH
+# ROOT = ROOT.relative_to(Path.cwd())  # relative
+
+import numpy as np
+import tensorflow as tf
+import torch
+import torch.nn as nn
+from tensorflow import keras
+
+from models.common import (C3, SPP, SPPF, Bottleneck, BottleneckCSP, C3x, Concat, Conv, CrossConv, DWConv,
+                           DWConvTranspose2d, Focus, autopad)
+from models.experimental import MixConv2d, attempt_load
+from models.yolo import Detect, Segment
+from utils.activations import SiLU
+from utils.general import LOGGER, make_divisible, print_args
+
+
+class TFBN(keras.layers.Layer):
+    # TensorFlow BatchNormalization wrapper
+    def __init__(self, w=None):
+        super().__init__()
+        self.bn = keras.layers.BatchNormalization(
+            beta_initializer=keras.initializers.Constant(w.bias.numpy()),
+            gamma_initializer=keras.initializers.Constant(w.weight.numpy()),
+            moving_mean_initializer=keras.initializers.Constant(w.running_mean.numpy()),
+            moving_variance_initializer=keras.initializers.Constant(w.running_var.numpy()),
+            epsilon=w.eps)
+
+    def call(self, inputs):
+        return self.bn(inputs)
+
+
+class TFPad(keras.layers.Layer):
+    # Pad inputs in spatial dimensions 1 and 2
+    def __init__(self, pad):
+        super().__init__()
+        if isinstance(pad, int):
+            self.pad = tf.constant([[0, 0], [pad, pad], [pad, pad], [0, 0]])
+        else:  # tuple/list
+            self.pad = tf.constant([[0, 0], [pad[0], pad[0]], [pad[1], pad[1]], [0, 0]])
+
+    def call(self, inputs):
+        return tf.pad(inputs, self.pad, mode='constant', constant_values=0)
+
+
+class TFConv(keras.layers.Layer):
+    # Standard convolution
+    def __init__(self, c1, c2, k=1, s=1, p=None, g=1, act=True, w=None):
+        # ch_in, ch_out, weights, kernel, stride, padding, groups
+        super().__init__()
+        assert g == 1, "TF v2.2 Conv2D does not support 'groups' argument"
+        # TensorFlow convolution padding is inconsistent with PyTorch (e.g. k=3 s=2 'SAME' padding)
+        # see https://stackoverflow.com/questions/52975843/comparing-conv2d-with-padding-between-tensorflow-and-pytorch
+        conv = keras.layers.Conv2D(
+            filters=c2,
+            kernel_size=k,
+            strides=s,
+            padding='SAME' if s == 1 else 'VALID',
+            use_bias=not hasattr(w, 'bn'),
+            kernel_initializer=keras.initializers.Constant(w.conv.weight.permute(2, 3, 1, 0).numpy()),
+            bias_initializer='zeros' if hasattr(w, 'bn') else keras.initializers.Constant(w.conv.bias.numpy()))
+        self.conv = conv if s == 1 else keras.Sequential([TFPad(autopad(k, p)), conv])
+        self.bn = TFBN(w.bn) if hasattr(w, 'bn') else tf.identity
+        self.act = activations(w.act) if act else tf.identity
+
+    def call(self, inputs):
+        return self.act(self.bn(self.conv(inputs)))
+
+
+class TFDWConv(keras.layers.Layer):
+    # Depthwise convolution
+    def __init__(self, c1, c2, k=1, s=1, p=None, act=True, w=None):
+        # ch_in, ch_out, weights, kernel, stride, padding, groups
+        super().__init__()
+        assert c2 % c1 == 0, f'TFDWConv() output={c2} must be a multiple of input={c1} channels'
+        conv = keras.layers.DepthwiseConv2D(
+            kernel_size=k,
+            depth_multiplier=c2 // c1,
+            strides=s,
+            padding='SAME' if s == 1 else 'VALID',
+            use_bias=not hasattr(w, 'bn'),
+            depthwise_initializer=keras.initializers.Constant(w.conv.weight.permute(2, 3, 1, 0).numpy()),
+            bias_initializer='zeros' if hasattr(w, 'bn') else keras.initializers.Constant(w.conv.bias.numpy()))
+        self.conv = conv if s == 1 else keras.Sequential([TFPad(autopad(k, p)), conv])
+        self.bn = TFBN(w.bn) if hasattr(w, 'bn') else tf.identity
+        self.act = activations(w.act) if act else tf.identity
+
+    def call(self, inputs):
+        return self.act(self.bn(self.conv(inputs)))
+
+
+class TFDWConvTranspose2d(keras.layers.Layer):
+    # Depthwise ConvTranspose2d
+    def __init__(self, c1, c2, k=1, s=1, p1=0, p2=0, w=None):
+        # ch_in, ch_out, weights, kernel, stride, padding, groups
+        super().__init__()
+        assert c1 == c2, f'TFDWConv() output={c2} must be equal to input={c1} channels'
+        assert k == 4 and p1 == 1, 'TFDWConv() only valid for k=4 and p1=1'
+        weight, bias = w.weight.permute(2, 3, 1, 0).numpy(), w.bias.numpy()
+        self.c1 = c1
+        self.conv = [
+            keras.layers.Conv2DTranspose(filters=1,
+                                         kernel_size=k,
+                                         strides=s,
+                                         padding='VALID',
+                                         output_padding=p2,
+                                         use_bias=True,
+                                         kernel_initializer=keras.initializers.Constant(weight[..., i:i + 1]),
+                                         bias_initializer=keras.initializers.Constant(bias[i])) for i in range(c1)]
+
+    def call(self, inputs):
+        return tf.concat([m(x) for m, x in zip(self.conv, tf.split(inputs, self.c1, 3))], 3)[:, 1:-1, 1:-1]
+
+
+class TFFocus(keras.layers.Layer):
+    # Focus wh information into c-space
+    def __init__(self, c1, c2, k=1, s=1, p=None, g=1, act=True, w=None):
+        # ch_in, ch_out, kernel, stride, padding, groups
+        super().__init__()
+        self.conv = TFConv(c1 * 4, c2, k, s, p, g, act, w.conv)
+
+    def call(self, inputs):  # x(b,w,h,c) -> y(b,w/2,h/2,4c)
+        # inputs = inputs / 255  # normalize 0-255 to 0-1
+        inputs = [inputs[:, ::2, ::2, :], inputs[:, 1::2, ::2, :], inputs[:, ::2, 1::2, :], inputs[:, 1::2, 1::2, :]]
+        return self.conv(tf.concat(inputs, 3))
+
+
+class TFBottleneck(keras.layers.Layer):
+    # Standard bottleneck
+    def __init__(self, c1, c2, shortcut=True, g=1, e=0.5, w=None):  # ch_in, ch_out, shortcut, groups, expansion
+        super().__init__()
+        c_ = int(c2 * e)  # hidden channels
+        self.cv1 = TFConv(c1, c_, 1, 1, w=w.cv1)
+        self.cv2 = TFConv(c_, c2, 3, 1, g=g, w=w.cv2)
+        self.add = shortcut and c1 == c2
+
+    def call(self, inputs):
+        return inputs + self.cv2(self.cv1(inputs)) if self.add else self.cv2(self.cv1(inputs))
+
+
+class TFCrossConv(keras.layers.Layer):
+    # Cross Convolution
+    def __init__(self, c1, c2, k=3, s=1, g=1, e=1.0, shortcut=False, w=None):
+        super().__init__()
+        c_ = int(c2 * e)  # hidden channels
+        self.cv1 = TFConv(c1, c_, (1, k), (1, s), w=w.cv1)
+        self.cv2 = TFConv(c_, c2, (k, 1), (s, 1), g=g, w=w.cv2)
+        self.add = shortcut and c1 == c2
+
+    def call(self, inputs):
+        return inputs + self.cv2(self.cv1(inputs)) if self.add else self.cv2(self.cv1(inputs))
+
+
+class TFConv2d(keras.layers.Layer):
+    # Substitution for PyTorch nn.Conv2D
+    def __init__(self, c1, c2, k, s=1, g=1, bias=True, w=None):
+        super().__init__()
+        assert g == 1, "TF v2.2 Conv2D does not support 'groups' argument"
+        self.conv = keras.layers.Conv2D(filters=c2,
+                                        kernel_size=k,
+                                        strides=s,
+                                        padding='VALID',
+                                        use_bias=bias,
+                                        kernel_initializer=keras.initializers.Constant(
+                                            w.weight.permute(2, 3, 1, 0).numpy()),
+                                        bias_initializer=keras.initializers.Constant(w.bias.numpy()) if bias else None)
+
+    def call(self, inputs):
+        return self.conv(inputs)
+
+
+class TFBottleneckCSP(keras.layers.Layer):
+    # CSP Bottleneck https://github.com/WongKinYiu/CrossStagePartialNetworks
+    def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5, w=None):
+        # ch_in, ch_out, number, shortcut, groups, expansion
+        super().__init__()
+        c_ = int(c2 * e)  # hidden channels
+        self.cv1 = TFConv(c1, c_, 1, 1, w=w.cv1)
+        self.cv2 = TFConv2d(c1, c_, 1, 1, bias=False, w=w.cv2)
+        self.cv3 = TFConv2d(c_, c_, 1, 1, bias=False, w=w.cv3)
+        self.cv4 = TFConv(2 * c_, c2, 1, 1, w=w.cv4)
+        self.bn = TFBN(w.bn)
+        self.act = lambda x: keras.activations.swish(x)
+        self.m = keras.Sequential([TFBottleneck(c_, c_, shortcut, g, e=1.0, w=w.m[j]) for j in range(n)])
+
+    def call(self, inputs):
+        y1 = self.cv3(self.m(self.cv1(inputs)))
+        y2 = self.cv2(inputs)
+        return self.cv4(self.act(self.bn(tf.concat((y1, y2), axis=3))))
+
+
+class TFC3(keras.layers.Layer):
+    # CSP Bottleneck with 3 convolutions
+    def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5, w=None):
+        # ch_in, ch_out, number, shortcut, groups, expansion
+        super().__init__()
+        c_ = int(c2 * e)  # hidden channels
+        self.cv1 = TFConv(c1, c_, 1, 1, w=w.cv1)
+        self.cv2 = TFConv(c1, c_, 1, 1, w=w.cv2)
+        self.cv3 = TFConv(2 * c_, c2, 1, 1, w=w.cv3)
+        self.m = keras.Sequential([TFBottleneck(c_, c_, shortcut, g, e=1.0, w=w.m[j]) for j in range(n)])
+
+    def call(self, inputs):
+        return self.cv3(tf.concat((self.m(self.cv1(inputs)), self.cv2(inputs)), axis=3))
+
+
+class TFC3x(keras.layers.Layer):
+    # 3 module with cross-convolutions
+    def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5, w=None):
+        # ch_in, ch_out, number, shortcut, groups, expansion
+        super().__init__()
+        c_ = int(c2 * e)  # hidden channels
+        self.cv1 = TFConv(c1, c_, 1, 1, w=w.cv1)
+        self.cv2 = TFConv(c1, c_, 1, 1, w=w.cv2)
+        self.cv3 = TFConv(2 * c_, c2, 1, 1, w=w.cv3)
+        self.m = keras.Sequential([
+            TFCrossConv(c_, c_, k=3, s=1, g=g, e=1.0, shortcut=shortcut, w=w.m[j]) for j in range(n)])
+
+    def call(self, inputs):
+        return self.cv3(tf.concat((self.m(self.cv1(inputs)), self.cv2(inputs)), axis=3))
+
+
+class TFSPP(keras.layers.Layer):
+    # Spatial pyramid pooling layer used in YOLOv3-SPP
+    def __init__(self, c1, c2, k=(5, 9, 13), w=None):
+        super().__init__()
+        c_ = c1 // 2  # hidden channels
+        self.cv1 = TFConv(c1, c_, 1, 1, w=w.cv1)
+        self.cv2 = TFConv(c_ * (len(k) + 1), c2, 1, 1, w=w.cv2)
+        self.m = [keras.layers.MaxPool2D(pool_size=x, strides=1, padding='SAME') for x in k]
+
+    def call(self, inputs):
+        x = self.cv1(inputs)
+        return self.cv2(tf.concat([x] + [m(x) for m in self.m], 3))
+
+
+class TFSPPF(keras.layers.Layer):
+    # Spatial pyramid pooling-Fast layer
+    def __init__(self, c1, c2, k=5, w=None):
+        super().__init__()
+        c_ = c1 // 2  # hidden channels
+        self.cv1 = TFConv(c1, c_, 1, 1, w=w.cv1)
+        self.cv2 = TFConv(c_ * 4, c2, 1, 1, w=w.cv2)
+        self.m = keras.layers.MaxPool2D(pool_size=k, strides=1, padding='SAME')
+
+    def call(self, inputs):
+        x = self.cv1(inputs)
+        y1 = self.m(x)
+        y2 = self.m(y1)
+        return self.cv2(tf.concat([x, y1, y2, self.m(y2)], 3))
+
+
+class TFDetect(keras.layers.Layer):
+    # TF YOLOv5 Detect layer
+    def __init__(self, nc=80, anchors=(), ch=(), imgsz=(640, 640), w=None):  # detection layer
+        super().__init__()
+        self.stride = tf.convert_to_tensor(w.stride.numpy(), dtype=tf.float32)
+        self.nc = nc  # number of classes
+        self.no = nc + 5  # number of outputs per anchor
+        self.nl = len(anchors)  # number of detection layers
+        self.na = len(anchors[0]) // 2  # number of anchors
+        self.grid = [tf.zeros(1)] * self.nl  # init grid
+        self.anchors = tf.convert_to_tensor(w.anchors.numpy(), dtype=tf.float32)
+        self.anchor_grid = tf.reshape(self.anchors * tf.reshape(self.stride, [self.nl, 1, 1]), [self.nl, 1, -1, 1, 2])
+        self.m = [TFConv2d(x, self.no * self.na, 1, w=w.m[i]) for i, x in enumerate(ch)]
+        self.training = False  # set to False after building model
+        self.imgsz = imgsz
+        for i in range(self.nl):
+            ny, nx = self.imgsz[0] // self.stride[i], self.imgsz[1] // self.stride[i]
+            self.grid[i] = self._make_grid(nx, ny)
+
+    def call(self, inputs):
+        z = []  # inference output
+        x = []
+        for i in range(self.nl):
+            x.append(self.m[i](inputs[i]))
+            # x(bs,20,20,255) to x(bs,3,20,20,85)
+            ny, nx = self.imgsz[0] // self.stride[i], self.imgsz[1] // self.stride[i]
+            x[i] = tf.reshape(x[i], [-1, ny * nx, self.na, self.no])
+
+            if not self.training:  # inference
+                y = x[i]
+                grid = tf.transpose(self.grid[i], [0, 2, 1, 3]) - 0.5
+                anchor_grid = tf.transpose(self.anchor_grid[i], [0, 2, 1, 3]) * 4
+                xy = (tf.sigmoid(y[..., 0:2]) * 2 + grid) * self.stride[i]  # xy
+                wh = tf.sigmoid(y[..., 2:4]) ** 2 * anchor_grid
+                # Normalize xywh to 0-1 to reduce calibration error
+                xy /= tf.constant([[self.imgsz[1], self.imgsz[0]]], dtype=tf.float32)
+                wh /= tf.constant([[self.imgsz[1], self.imgsz[0]]], dtype=tf.float32)
+                y = tf.concat([xy, wh, tf.sigmoid(y[..., 4:5 + self.nc]), y[..., 5 + self.nc:]], -1)
+                z.append(tf.reshape(y, [-1, self.na * ny * nx, self.no]))
+
+        return tf.transpose(x, [0, 2, 1, 3]) if self.training else (tf.concat(z, 1), )
+
+    @staticmethod
+    def _make_grid(nx=20, ny=20):
+        # yv, xv = torch.meshgrid([torch.arange(ny), torch.arange(nx)])
+        # return torch.stack((xv, yv), 2).view((1, 1, ny, nx, 2)).float()
+        xv, yv = tf.meshgrid(tf.range(nx), tf.range(ny))
+        return tf.cast(tf.reshape(tf.stack([xv, yv], 2), [1, 1, ny * nx, 2]), dtype=tf.float32)
+
+
+class TFSegment(TFDetect):
+    # YOLOv5 Segment head for segmentation models
+    def __init__(self, nc=80, anchors=(), nm=32, npr=256, ch=(), imgsz=(640, 640), w=None):
+        super().__init__(nc, anchors, ch, imgsz, w)
+        self.nm = nm  # number of masks
+        self.npr = npr  # number of protos
+        self.no = 5 + nc + self.nm  # number of outputs per anchor
+        self.m = [TFConv2d(x, self.no * self.na, 1, w=w.m[i]) for i, x in enumerate(ch)]  # output conv
+        self.proto = TFProto(ch[0], self.npr, self.nm, w=w.proto)  # protos
+        self.detect = TFDetect.call
+
+    def call(self, x):
+        p = self.proto(x[0])
+        # p = TFUpsample(None, scale_factor=4, mode='nearest')(self.proto(x[0]))  # (optional) full-size protos
+        p = tf.transpose(p, [0, 3, 1, 2])  # from shape(1,160,160,32) to shape(1,32,160,160)
+        x = self.detect(self, x)
+        return (x, p) if self.training else (x[0], p)
+
+
+class TFProto(keras.layers.Layer):
+
+    def __init__(self, c1, c_=256, c2=32, w=None):
+        super().__init__()
+        self.cv1 = TFConv(c1, c_, k=3, w=w.cv1)
+        self.upsample = TFUpsample(None, scale_factor=2, mode='nearest')
+        self.cv2 = TFConv(c_, c_, k=3, w=w.cv2)
+        self.cv3 = TFConv(c_, c2, w=w.cv3)
+
+    def call(self, inputs):
+        return self.cv3(self.cv2(self.upsample(self.cv1(inputs))))
+
+
+class TFUpsample(keras.layers.Layer):
+    # TF version of torch.nn.Upsample()
+    def __init__(self, size, scale_factor, mode, w=None):  # warning: all arguments needed including 'w'
+        super().__init__()
+        assert scale_factor % 2 == 0, 'scale_factor must be multiple of 2'
+        self.upsample = lambda x: tf.image.resize(x, (x.shape[1] * scale_factor, x.shape[2] * scale_factor), mode)
+        # self.upsample = keras.layers.UpSampling2D(size=scale_factor, interpolation=mode)
+        # with default arguments: align_corners=False, half_pixel_centers=False
+        # self.upsample = lambda x: tf.raw_ops.ResizeNearestNeighbor(images=x,
+        #                                                            size=(x.shape[1] * 2, x.shape[2] * 2))
+
+    def call(self, inputs):
+        return self.upsample(inputs)
+
+
+class TFConcat(keras.layers.Layer):
+    # TF version of torch.concat()
+    def __init__(self, dimension=1, w=None):
+        super().__init__()
+        assert dimension == 1, 'convert only NCHW to NHWC concat'
+        self.d = 3
+
+    def call(self, inputs):
+        return tf.concat(inputs, self.d)
+
+
+def parse_model(d, ch, model, imgsz):  # model_dict, input_channels(3)
+    LOGGER.info(f"\n{'':>3}{'from':>18}{'n':>3}{'params':>10}  {'module':<40}{'arguments':<30}")
+    anchors, nc, gd, gw = d['anchors'], d['nc'], d['depth_multiple'], d['width_multiple']
+    na = (len(anchors[0]) // 2) if isinstance(anchors, list) else anchors  # number of anchors
+    no = na * (nc + 5)  # number of outputs = anchors * (classes + 5)
+
+    layers, save, c2 = [], [], ch[-1]  # layers, savelist, ch out
+    for i, (f, n, m, args) in enumerate(d['backbone'] + d['head']):  # from, number, module, args
+        m_str = m
+        m = eval(m) if isinstance(m, str) else m  # eval strings
+        for j, a in enumerate(args):
+            try:
+                args[j] = eval(a) if isinstance(a, str) else a  # eval strings
+            except NameError:
+                pass
+
+        n = max(round(n * gd), 1) if n > 1 else n  # depth gain
+        if m in [
+                nn.Conv2d, Conv, DWConv, DWConvTranspose2d, Bottleneck, SPP, SPPF, MixConv2d, Focus, CrossConv,
+                BottleneckCSP, C3, C3x]:
+            c1, c2 = ch[f], args[0]
+            c2 = make_divisible(c2 * gw, 8) if c2 != no else c2
+
+            args = [c1, c2, *args[1:]]
+            if m in [BottleneckCSP, C3, C3x]:
+                args.insert(2, n)
+                n = 1
+        elif m is nn.BatchNorm2d:
+            args = [ch[f]]
+        elif m is Concat:
+            c2 = sum(ch[-1 if x == -1 else x + 1] for x in f)
+        elif m in [Detect, Segment]:
+            args.append([ch[x + 1] for x in f])
+            if isinstance(args[1], int):  # number of anchors
+                args[1] = [list(range(args[1] * 2))] * len(f)
+            if m is Segment:
+                args[3] = make_divisible(args[3] * gw, 8)
+            args.append(imgsz)
+        else:
+            c2 = ch[f]
+
+        tf_m = eval('TF' + m_str.replace('nn.', ''))
+        m_ = keras.Sequential([tf_m(*args, w=model.model[i][j]) for j in range(n)]) if n > 1 \
+            else tf_m(*args, w=model.model[i])  # module
+
+        torch_m_ = nn.Sequential(*(m(*args) for _ in range(n))) if n > 1 else m(*args)  # module
+        t = str(m)[8:-2].replace('__main__.', '')  # module type
+        np = sum(x.numel() for x in torch_m_.parameters())  # number params
+        m_.i, m_.f, m_.type, m_.np = i, f, t, np  # attach index, 'from' index, type, number params
+        LOGGER.info(f'{i:>3}{str(f):>18}{str(n):>3}{np:>10}  {t:<40}{str(args):<30}')  # print
+        save.extend(x % i for x in ([f] if isinstance(f, int) else f) if x != -1)  # append to savelist
+        layers.append(m_)
+        ch.append(c2)
+    return keras.Sequential(layers), sorted(save)
+
+
+class TFModel:
+    # TF YOLOv5 model
+    def __init__(self, cfg='yolov5s.yaml', ch=3, nc=None, model=None, imgsz=(640, 640)):  # model, channels, classes
+        super().__init__()
+        if isinstance(cfg, dict):
+            self.yaml = cfg  # model dict
+        else:  # is *.yaml
+            import yaml  # for torch hub
+            self.yaml_file = Path(cfg).name
+            with open(cfg) as f:
+                self.yaml = yaml.load(f, Loader=yaml.FullLoader)  # model dict
+
+        # Define model
+        if nc and nc != self.yaml['nc']:
+            LOGGER.info(f"Overriding {cfg} nc={self.yaml['nc']} with nc={nc}")
+            self.yaml['nc'] = nc  # override yaml value
+        self.model, self.savelist = parse_model(deepcopy(self.yaml), ch=[ch], model=model, imgsz=imgsz)
+
+    def predict(self,
+                inputs,
+                tf_nms=False,
+                agnostic_nms=False,
+                topk_per_class=100,
+                topk_all=100,
+                iou_thres=0.45,
+                conf_thres=0.25):
+        y = []  # outputs
+        x = inputs
+        for m in self.model.layers:
+            if m.f != -1:  # if not from previous layer
+                x = y[m.f] if isinstance(m.f, int) else [x if j == -1 else y[j] for j in m.f]  # from earlier layers
+
+            x = m(x)  # run
+            y.append(x if m.i in self.savelist else None)  # save output
+
+        # Add TensorFlow NMS
+        if tf_nms:
+            boxes = self._xywh2xyxy(x[0][..., :4])
+            probs = x[0][:, :, 4:5]
+            classes = x[0][:, :, 5:]
+            scores = probs * classes
+            if agnostic_nms:
+                nms = AgnosticNMS()((boxes, classes, scores), topk_all, iou_thres, conf_thres)
+            else:
+                boxes = tf.expand_dims(boxes, 2)
+                nms = tf.image.combined_non_max_suppression(boxes,
+                                                            scores,
+                                                            topk_per_class,
+                                                            topk_all,
+                                                            iou_thres,
+                                                            conf_thres,
+                                                            clip_boxes=False)
+            return (nms, )
+        return x  # output [1,6300,85] = [xywh, conf, class0, class1, ...]
+        # x = x[0]  # [x(1,6300,85), ...] to x(6300,85)
+        # xywh = x[..., :4]  # x(6300,4) boxes
+        # conf = x[..., 4:5]  # x(6300,1) confidences
+        # cls = tf.reshape(tf.cast(tf.argmax(x[..., 5:], axis=1), tf.float32), (-1, 1))  # x(6300,1)  classes
+        # return tf.concat([conf, cls, xywh], 1)
+
+    @staticmethod
+    def _xywh2xyxy(xywh):
+        # Convert nx4 boxes from [x, y, w, h] to [x1, y1, x2, y2] where xy1=top-left, xy2=bottom-right
+        x, y, w, h = tf.split(xywh, num_or_size_splits=4, axis=-1)
+        return tf.concat([x - w / 2, y - h / 2, x + w / 2, y + h / 2], axis=-1)
+
+
+class AgnosticNMS(keras.layers.Layer):
+    # TF Agnostic NMS
+    def call(self, input, topk_all, iou_thres, conf_thres):
+        # wrap map_fn to avoid TypeSpec related error https://stackoverflow.com/a/65809989/3036450
+        return tf.map_fn(lambda x: self._nms(x, topk_all, iou_thres, conf_thres),
+                         input,
+                         fn_output_signature=(tf.float32, tf.float32, tf.float32, tf.int32),
+                         name='agnostic_nms')
+
+    @staticmethod
+    def _nms(x, topk_all=100, iou_thres=0.45, conf_thres=0.25):  # agnostic NMS
+        boxes, classes, scores = x
+        class_inds = tf.cast(tf.argmax(classes, axis=-1), tf.float32)
+        scores_inp = tf.reduce_max(scores, -1)
+        selected_inds = tf.image.non_max_suppression(boxes,
+                                                     scores_inp,
+                                                     max_output_size=topk_all,
+                                                     iou_threshold=iou_thres,
+                                                     score_threshold=conf_thres)
+        selected_boxes = tf.gather(boxes, selected_inds)
+        padded_boxes = tf.pad(selected_boxes,
+                              paddings=[[0, topk_all - tf.shape(selected_boxes)[0]], [0, 0]],
+                              mode='CONSTANT',
+                              constant_values=0.0)
+        selected_scores = tf.gather(scores_inp, selected_inds)
+        padded_scores = tf.pad(selected_scores,
+                               paddings=[[0, topk_all - tf.shape(selected_boxes)[0]]],
+                               mode='CONSTANT',
+                               constant_values=-1.0)
+        selected_classes = tf.gather(class_inds, selected_inds)
+        padded_classes = tf.pad(selected_classes,
+                                paddings=[[0, topk_all - tf.shape(selected_boxes)[0]]],
+                                mode='CONSTANT',
+                                constant_values=-1.0)
+        valid_detections = tf.shape(selected_inds)[0]
+        return padded_boxes, padded_scores, padded_classes, valid_detections
+
+
+def activations(act=nn.SiLU):
+    # Returns TF activation from input PyTorch activation
+    if isinstance(act, nn.LeakyReLU):
+        return lambda x: keras.activations.relu(x, alpha=0.1)
+    elif isinstance(act, nn.Hardswish):
+        return lambda x: x * tf.nn.relu6(x + 3) * 0.166666667
+    elif isinstance(act, (nn.SiLU, SiLU)):
+        return lambda x: keras.activations.swish(x)
+    else:
+        raise Exception(f'no matching TensorFlow activation found for PyTorch activation {act}')
+
+
+def representative_dataset_gen(dataset, ncalib=100):
+    # Representative dataset generator for use with converter.representative_dataset, returns a generator of np arrays
+    for n, (path, img, im0s, vid_cap, string) in enumerate(dataset):
+        im = np.transpose(img, [1, 2, 0])
+        im = np.expand_dims(im, axis=0).astype(np.float32)
+        im /= 255
+        yield [im]
+        if n >= ncalib:
+            break
+
+
+def run(
+        weights=ROOT / 'yolov5s.pt',  # weights path
+        imgsz=(640, 640),  # inference size h,w
+        batch_size=1,  # batch size
+        dynamic=False,  # dynamic batch size
+):
+    # PyTorch model
+    im = torch.zeros((batch_size, 3, *imgsz))  # BCHW image
+    model = attempt_load(weights, device=torch.device('cpu'), inplace=True, fuse=False)
+    _ = model(im)  # inference
+    model.info()
+
+    # TensorFlow model
+    im = tf.zeros((batch_size, *imgsz, 3))  # BHWC image
+    tf_model = TFModel(cfg=model.yaml, model=model, nc=model.nc, imgsz=imgsz)
+    _ = tf_model.predict(im)  # inference
+
+    # Keras model
+    im = keras.Input(shape=(*imgsz, 3), batch_size=None if dynamic else batch_size)
+    keras_model = keras.Model(inputs=im, outputs=tf_model.predict(im))
+    keras_model.summary()
+
+    LOGGER.info('PyTorch, TensorFlow and Keras models successfully verified.\nUse export.py for TF model export.')
+
+
+def parse_opt():
+    parser = argparse.ArgumentParser()
+    parser.add_argument('--weights', type=str, default=ROOT / 'yolov5s.pt', help='weights path')
+    parser.add_argument('--imgsz', '--img', '--img-size', nargs='+', type=int, default=[640], help='inference size h,w')
+    parser.add_argument('--batch-size', type=int, default=1, help='batch size')
+    parser.add_argument('--dynamic', action='store_true', help='dynamic batch size')
+    opt = parser.parse_args()
+    opt.imgsz *= 2 if len(opt.imgsz) == 1 else 1  # expand
+    print_args(vars(opt))
+    return opt
+
+
+def main(opt):
+    run(**vars(opt))
+
+
+if __name__ == '__main__':
+    opt = parse_opt()
+    main(opt)

models/yolo.py

@@ -0,0 +1,391 @@
+# YOLOv5 🚀 by Ultralytics, AGPL-3.0 license
+"""
+YOLO-specific modules
+
+Usage:
+    $ python models/yolo.py --cfg yolov5s.yaml
+"""
+
+import argparse
+import contextlib
+import os
+import platform
+import sys
+from copy import deepcopy
+from pathlib import Path
+
+FILE = Path(__file__).resolve()
+ROOT = FILE.parents[1]  # YOLOv5 root directory
+if str(ROOT) not in sys.path:
+    sys.path.append(str(ROOT))  # add ROOT to PATH
+if platform.system() != 'Windows':
+    ROOT = Path(os.path.relpath(ROOT, Path.cwd()))  # relative
+
+from models.common import *  # noqa
+from models.experimental import *  # noqa
+from utils.autoanchor import check_anchor_order
+from utils.general import LOGGER, check_version, check_yaml, make_divisible, print_args
+from utils.plots import feature_visualization
+from utils.torch_utils import (fuse_conv_and_bn, initialize_weights, model_info, profile, scale_img, select_device,
+                               time_sync)
+
+try:
+    import thop  # for FLOPs computation
+except ImportError:
+    thop = None
+
+
+class Detect(nn.Module):
+    # YOLOv5 Detect head for detection models
+    stride = None  # strides computed during build
+    dynamic = False  # force grid reconstruction
+    export = False  # export mode
+
+    def __init__(self, nc=80, anchors=(), ch=(), inplace=True):  # detection layer
+        super().__init__()
+        self.nc = nc  # number of classes
+        self.no = nc + 5  # number of outputs per anchor
+        self.nl = len(anchors)  # number of detection layers
+        self.na = len(anchors[0]) // 2  # number of anchors
+        self.grid = [torch.empty(0) for _ in range(self.nl)]  # init grid
+        self.anchor_grid = [torch.empty(0) for _ in range(self.nl)]  # init anchor grid
+        self.register_buffer('anchors', torch.tensor(anchors).float().view(self.nl, -1, 2))  # shape(nl,na,2)
+        self.m = nn.ModuleList(nn.Conv2d(x, self.no * self.na, 1) for x in ch)  # output conv
+        self.inplace = inplace  # use inplace ops (e.g. slice assignment)
+
+    def forward(self, x):
+        z = []  # inference output
+        for i in range(self.nl):
+            x[i] = self.m[i](x[i])  # conv
+            bs, _, ny, nx = x[i].shape  # x(bs,255,20,20) to x(bs,3,20,20,85)
+            x[i] = x[i].view(bs, self.na, self.no, ny, nx).permute(0, 1, 3, 4, 2).contiguous()
+
+            if not self.training:  # inference
+                if self.dynamic or self.grid[i].shape[2:4] != x[i].shape[2:4]:
+                    self.grid[i], self.anchor_grid[i] = self._make_grid(nx, ny, i)
+
+                if isinstance(self, Segment):  # (boxes + masks)
+                    xy, wh, conf, mask = x[i].split((2, 2, self.nc + 1, self.no - self.nc - 5), 4)
+                    xy = (xy.sigmoid() * 2 + self.grid[i]) * self.stride[i]  # xy
+                    wh = (wh.sigmoid() * 2) ** 2 * self.anchor_grid[i]  # wh
+                    y = torch.cat((xy, wh, conf.sigmoid(), mask), 4)
+                else:  # Detect (boxes only)
+                    xy, wh, conf = x[i].sigmoid().split((2, 2, self.nc + 1), 4)
+                    xy = (xy * 2 + self.grid[i]) * self.stride[i]  # xy
+                    wh = (wh * 2) ** 2 * self.anchor_grid[i]  # wh
+                    y = torch.cat((xy, wh, conf), 4)
+                z.append(y.view(bs, self.na * nx * ny, self.no))
+
+        return x if self.training else (torch.cat(z, 1), ) if self.export else (torch.cat(z, 1), x)
+
+    def _make_grid(self, nx=20, ny=20, i=0, torch_1_10=check_version(torch.__version__, '1.10.0')):
+        d = self.anchors[i].device
+        t = self.anchors[i].dtype
+        shape = 1, self.na, ny, nx, 2  # grid shape
+        y, x = torch.arange(ny, device=d, dtype=t), torch.arange(nx, device=d, dtype=t)
+        yv, xv = torch.meshgrid(y, x, indexing='ij') if torch_1_10 else torch.meshgrid(y, x)  # torch>=0.7 compatibility
+        grid = torch.stack((xv, yv), 2).expand(shape) - 0.5  # add grid offset, i.e. y = 2.0 * x - 0.5
+        anchor_grid = (self.anchors[i] * self.stride[i]).view((1, self.na, 1, 1, 2)).expand(shape)
+        return grid, anchor_grid
+
+
+class Segment(Detect):
+    # YOLOv5 Segment head for segmentation models
+    def __init__(self, nc=80, anchors=(), nm=32, npr=256, ch=(), inplace=True):
+        super().__init__(nc, anchors, ch, inplace)
+        self.nm = nm  # number of masks
+        self.npr = npr  # number of protos
+        self.no = 5 + nc + self.nm  # number of outputs per anchor
+        self.m = nn.ModuleList(nn.Conv2d(x, self.no * self.na, 1) for x in ch)  # output conv
+        self.proto = Proto(ch[0], self.npr, self.nm)  # protos
+        self.detect = Detect.forward
+
+    def forward(self, x):
+        p = self.proto(x[0])
+        x = self.detect(self, x)
+        return (x, p) if self.training else (x[0], p) if self.export else (x[0], p, x[1])
+
+
+class BaseModel(nn.Module):
+    # YOLOv5 base model
+    def forward(self, x, profile=False, visualize=False):
+        return self._forward_once(x, profile, visualize)  # single-scale inference, train
+
+    def _forward_once(self, x, profile=False, visualize=False):
+        y, dt = [], []  # outputs
+        for m in self.model:
+            if m.f != -1:  # if not from previous layer
+                x = y[m.f] if isinstance(m.f, int) else [x if j == -1 else y[j] for j in m.f]  # from earlier layers
+            if profile:
+                self._profile_one_layer(m, x, dt)
+            x = m(x)  # run
+            y.append(x if m.i in self.save else None)  # save output
+            if visualize:
+                feature_visualization(x, m.type, m.i, save_dir=visualize)
+        return x
+
+    def _profile_one_layer(self, m, x, dt):
+        c = m == self.model[-1]  # is final layer, copy input as inplace fix
+        o = thop.profile(m, inputs=(x.copy() if c else x, ), verbose=False)[0] / 1E9 * 2 if thop else 0  # FLOPs
+        t = time_sync()
+        for _ in range(10):
+            m(x.copy() if c else x)
+        dt.append((time_sync() - t) * 100)
+        if m == self.model[0]:
+            LOGGER.info(f"{'time (ms)':>10s} {'GFLOPs':>10s} {'params':>10s}  module")
+        LOGGER.info(f'{dt[-1]:10.2f} {o:10.2f} {m.np:10.0f}  {m.type}')
+        if c:
+            LOGGER.info(f"{sum(dt):10.2f} {'-':>10s} {'-':>10s}  Total")
+
+    def fuse(self):  # fuse model Conv2d() + BatchNorm2d() layers
+        LOGGER.info('Fusing layers... ')
+        for m in self.model.modules():
+            if isinstance(m, (Conv, DWConv)) and hasattr(m, 'bn'):
+                m.conv = fuse_conv_and_bn(m.conv, m.bn)  # update conv
+                delattr(m, 'bn')  # remove batchnorm
+                m.forward = m.forward_fuse  # update forward
+        self.info()
+        return self
+
+    def info(self, verbose=False, img_size=640):  # print model information
+        model_info(self, verbose, img_size)
+
+    def _apply(self, fn):
+        # Apply to(), cpu(), cuda(), half() to model tensors that are not parameters or registered buffers
+        self = super()._apply(fn)
+        m = self.model[-1]  # Detect()
+        if isinstance(m, (Detect, Segment)):
+            m.stride = fn(m.stride)
+            m.grid = list(map(fn, m.grid))
+            if isinstance(m.anchor_grid, list):
+                m.anchor_grid = list(map(fn, m.anchor_grid))
+        return self
+
+
+class DetectionModel(BaseModel):
+    # YOLOv5 detection model
+    def __init__(self, cfg='yolov5s.yaml', ch=3, nc=None, anchors=None):  # model, input channels, number of classes
+        super().__init__()
+        if isinstance(cfg, dict):
+            self.yaml = cfg  # model dict
+        else:  # is *.yaml
+            import yaml  # for torch hub
+            self.yaml_file = Path(cfg).name
+            with open(cfg, encoding='ascii', errors='ignore') as f:
+                self.yaml = yaml.safe_load(f)  # model dict
+
+        # Define model
+        ch = self.yaml['ch'] = self.yaml.get('ch', ch)  # input channels
+        if nc and nc != self.yaml['nc']:
+            LOGGER.info(f"Overriding model.yaml nc={self.yaml['nc']} with nc={nc}")
+            self.yaml['nc'] = nc  # override yaml value
+        if anchors:
+            LOGGER.info(f'Overriding model.yaml anchors with anchors={anchors}')
+            self.yaml['anchors'] = round(anchors)  # override yaml value
+        self.model, self.save = parse_model(deepcopy(self.yaml), ch=[ch])  # model, savelist
+        self.names = [str(i) for i in range(self.yaml['nc'])]  # default names
+        self.inplace = self.yaml.get('inplace', True)
+
+        # Build strides, anchors
+        m = self.model[-1]  # Detect()
+        if isinstance(m, (Detect, Segment)):
+            s = 256  # 2x min stride
+            m.inplace = self.inplace
+            forward = lambda x: self.forward(x)[0] if isinstance(m, Segment) else self.forward(x)
+            m.stride = torch.tensor([s / x.shape[-2] for x in forward(torch.zeros(1, ch, s, s))])  # forward
+            check_anchor_order(m)
+            m.anchors /= m.stride.view(-1, 1, 1)
+            self.stride = m.stride
+            self._initialize_biases()  # only run once
+
+        # Init weights, biases
+        initialize_weights(self)
+        self.info()
+        LOGGER.info('')
+
+    def forward(self, x, augment=False, profile=False, visualize=False):
+        if augment:
+            return self._forward_augment(x)  # augmented inference, None
+        return self._forward_once(x, profile, visualize)  # single-scale inference, train
+
+    def _forward_augment(self, x):
+        img_size = x.shape[-2:]  # height, width
+        s = [1, 0.83, 0.67]  # scales
+        f = [None, 3, None]  # flips (2-ud, 3-lr)
+        y = []  # outputs
+        for si, fi in zip(s, f):
+            xi = scale_img(x.flip(fi) if fi else x, si, gs=int(self.stride.max()))
+            yi = self._forward_once(xi)[0]  # forward
+            # cv2.imwrite(f'img_{si}.jpg', 255 * xi[0].cpu().numpy().transpose((1, 2, 0))[:, :, ::-1])  # save
+            yi = self._descale_pred(yi, fi, si, img_size)
+            y.append(yi)
+        y = self._clip_augmented(y)  # clip augmented tails
+        return torch.cat(y, 1), None  # augmented inference, train
+
+    def _descale_pred(self, p, flips, scale, img_size):
+        # de-scale predictions following augmented inference (inverse operation)
+        if self.inplace:
+            p[..., :4] /= scale  # de-scale
+            if flips == 2:
+                p[..., 1] = img_size[0] - p[..., 1]  # de-flip ud
+            elif flips == 3:
+                p[..., 0] = img_size[1] - p[..., 0]  # de-flip lr
+        else:
+            x, y, wh = p[..., 0:1] / scale, p[..., 1:2] / scale, p[..., 2:4] / scale  # de-scale
+            if flips == 2:
+                y = img_size[0] - y  # de-flip ud
+            elif flips == 3:
+                x = img_size[1] - x  # de-flip lr
+            p = torch.cat((x, y, wh, p[..., 4:]), -1)
+        return p
+
+    def _clip_augmented(self, y):
+        # Clip YOLOv5 augmented inference tails
+        nl = self.model[-1].nl  # number of detection layers (P3-P5)
+        g = sum(4 ** x for x in range(nl))  # grid points
+        e = 1  # exclude layer count
+        i = (y[0].shape[1] // g) * sum(4 ** x for x in range(e))  # indices
+        y[0] = y[0][:, :-i]  # large
+        i = (y[-1].shape[1] // g) * sum(4 ** (nl - 1 - x) for x in range(e))  # indices
+        y[-1] = y[-1][:, i:]  # small
+        return y
+
+    def _initialize_biases(self, cf=None):  # initialize biases into Detect(), cf is class frequency
+        # https://arxiv.org/abs/1708.02002 section 3.3
+        # cf = torch.bincount(torch.tensor(np.concatenate(dataset.labels, 0)[:, 0]).long(), minlength=nc) + 1.
+        m = self.model[-1]  # Detect() module
+        for mi, s in zip(m.m, m.stride):  # from
+            b = mi.bias.view(m.na, -1)  # conv.bias(255) to (3,85)
+            b.data[:, 4] += math.log(8 / (640 / s) ** 2)  # obj (8 objects per 640 image)
+            b.data[:, 5:5 + m.nc] += math.log(0.6 / (m.nc - 0.99999)) if cf is None else torch.log(cf / cf.sum())  # cls
+            mi.bias = torch.nn.Parameter(b.view(-1), requires_grad=True)
+
+
+Model = DetectionModel  # retain YOLOv5 'Model' class for backwards compatibility
+
+
+class SegmentationModel(DetectionModel):
+    # YOLOv5 segmentation model
+    def __init__(self, cfg='yolov5s-seg.yaml', ch=3, nc=None, anchors=None):
+        super().__init__(cfg, ch, nc, anchors)
+
+
+class ClassificationModel(BaseModel):
+    # YOLOv5 classification model
+    def __init__(self, cfg=None, model=None, nc=1000, cutoff=10):  # yaml, model, number of classes, cutoff index
+        super().__init__()
+        self._from_detection_model(model, nc, cutoff) if model is not None else self._from_yaml(cfg)
+
+    def _from_detection_model(self, model, nc=1000, cutoff=10):
+        # Create a YOLOv5 classification model from a YOLOv5 detection model
+        if isinstance(model, DetectMultiBackend):
+            model = model.model  # unwrap DetectMultiBackend
+        model.model = model.model[:cutoff]  # backbone
+        m = model.model[-1]  # last layer
+        ch = m.conv.in_channels if hasattr(m, 'conv') else m.cv1.conv.in_channels  # ch into module
+        c = Classify(ch, nc)  # Classify()
+        c.i, c.f, c.type = m.i, m.f, 'models.common.Classify'  # index, from, type
+        model.model[-1] = c  # replace
+        self.model = model.model
+        self.stride = model.stride
+        self.save = []
+        self.nc = nc
+
+    def _from_yaml(self, cfg):
+        # Create a YOLOv5 classification model from a *.yaml file
+        self.model = None
+
+
+def parse_model(d, ch):  # model_dict, input_channels(3)
+    # Parse a YOLOv5 model.yaml dictionary
+    LOGGER.info(f"\n{'':>3}{'from':>18}{'n':>3}{'params':>10}  {'module':<40}{'arguments':<30}")
+    anchors, nc, gd, gw, act = d['anchors'], d['nc'], d['depth_multiple'], d['width_multiple'], d.get('activation')
+    if act:
+        Conv.default_act = eval(act)  # redefine default activation, i.e. Conv.default_act = nn.SiLU()
+        LOGGER.info(f"{colorstr('activation:')} {act}")  # print
+    na = (len(anchors[0]) // 2) if isinstance(anchors, list) else anchors  # number of anchors
+    no = na * (nc + 5)  # number of outputs = anchors * (classes + 5)
+
+    layers, save, c2 = [], [], ch[-1]  # layers, savelist, ch out
+    for i, (f, n, m, args) in enumerate(d['backbone'] + d['head']):  # from, number, module, args
+        m = eval(m) if isinstance(m, str) else m  # eval strings
+        for j, a in enumerate(args):
+            with contextlib.suppress(NameError):
+                args[j] = eval(a) if isinstance(a, str) else a  # eval strings
+
+        n = n_ = max(round(n * gd), 1) if n > 1 else n  # depth gain
+        if m in {
+                Conv, GhostConv, Bottleneck, GhostBottleneck, SPP, SPPF, DWConv, MixConv2d, Focus, CrossConv,
+                BottleneckCSP, C3, C3TR, C3SPP, C3Ghost, nn.ConvTranspose2d, DWConvTranspose2d, C3x}:
+            c1, c2 = ch[f], args[0]
+            if c2 != no:  # if not output
+                c2 = make_divisible(c2 * gw, 8)
+
+            args = [c1, c2, *args[1:]]
+            if m in {BottleneckCSP, C3, C3TR, C3Ghost, C3x}:
+                args.insert(2, n)  # number of repeats
+                n = 1
+        elif m is nn.BatchNorm2d:
+            args = [ch[f]]
+        elif m is Concat:
+            c2 = sum(ch[x] for x in f)
+        # TODO: channel, gw, gd
+        elif m in {Detect, Segment}:
+            args.append([ch[x] for x in f])
+            if isinstance(args[1], int):  # number of anchors
+                args[1] = [list(range(args[1] * 2))] * len(f)
+            if m is Segment:
+                args[3] = make_divisible(args[3] * gw, 8)
+        elif m is Contract:
+            c2 = ch[f] * args[0] ** 2
+        elif m is Expand:
+            c2 = ch[f] // args[0] ** 2
+        else:
+            c2 = ch[f]
+
+        m_ = nn.Sequential(*(m(*args) for _ in range(n))) if n > 1 else m(*args)  # module
+        t = str(m)[8:-2].replace('__main__.', '')  # module type
+        np = sum(x.numel() for x in m_.parameters())  # number params
+        m_.i, m_.f, m_.type, m_.np = i, f, t, np  # attach index, 'from' index, type, number params
+        LOGGER.info(f'{i:>3}{str(f):>18}{n_:>3}{np:10.0f}  {t:<40}{str(args):<30}')  # print
+        save.extend(x % i for x in ([f] if isinstance(f, int) else f) if x != -1)  # append to savelist
+        layers.append(m_)
+        if i == 0:
+            ch = []
+        ch.append(c2)
+    return nn.Sequential(*layers), sorted(save)
+
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser()
+    parser.add_argument('--cfg', type=str, default='yolov5s.yaml', help='model.yaml')
+    parser.add_argument('--batch-size', type=int, default=1, help='total batch size for all GPUs')
+    parser.add_argument('--device', default='', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')
+    parser.add_argument('--profile', action='store_true', help='profile model speed')
+    parser.add_argument('--line-profile', action='store_true', help='profile model speed layer by layer')
+    parser.add_argument('--test', action='store_true', help='test all yolo*.yaml')
+    opt = parser.parse_args()
+    opt.cfg = check_yaml(opt.cfg)  # check YAML
+    print_args(vars(opt))
+    device = select_device(opt.device)
+
+    # Create model
+    im = torch.rand(opt.batch_size, 3, 640, 640).to(device)
+    model = Model(opt.cfg).to(device)
+
+    # Options
+    if opt.line_profile:  # profile layer by layer
+        model(im, profile=True)
+
+    elif opt.profile:  # profile forward-backward
+        results = profile(input=im, ops=[model], n=3)
+
+    elif opt.test:  # test all models
+        for cfg in Path(ROOT / 'models').rglob('yolo*.yaml'):
+            try:
+                _ = Model(cfg)
+            except Exception as e:
+                print(f'Error in {cfg}: {e}')
+
+    else:  # report fused model summary
+        model.fuse()

models/yolov5l.yaml

@@ -0,0 +1,48 @@
+# YOLOv5 🚀 by Ultralytics, AGPL-3.0 license
+
+# Parameters
+nc: 80  # number of classes
+depth_multiple: 1.0  # model depth multiple
+width_multiple: 1.0  # layer channel multiple
+anchors:
+  - [10,13, 16,30, 33,23]  # P3/8
+  - [30,61, 62,45, 59,119]  # P4/16
+  - [116,90, 156,198, 373,326]  # P5/32
+
+# YOLOv5 v6.0 backbone
+backbone:
+  # [from, number, module, args]
+  [[-1, 1, Conv, [64, 6, 2, 2]],  # 0-P1/2
+   [-1, 1, Conv, [128, 3, 2]],  # 1-P2/4
+   [-1, 3, C3, [128]],
+   [-1, 1, Conv, [256, 3, 2]],  # 3-P3/8
+   [-1, 6, C3, [256]],
+   [-1, 1, Conv, [512, 3, 2]],  # 5-P4/16
+   [-1, 9, C3, [512]],
+   [-1, 1, Conv, [1024, 3, 2]],  # 7-P5/32
+   [-1, 3, C3, [1024]],
+   [-1, 1, SPPF, [1024, 5]],  # 9
+  ]
+
+# YOLOv5 v6.0 head
+head:
+  [[-1, 1, Conv, [512, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 6], 1, Concat, [1]],  # cat backbone P4
+   [-1, 3, C3, [512, False]],  # 13
+
+   [-1, 1, Conv, [256, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 4], 1, Concat, [1]],  # cat backbone P3
+   [-1, 3, C3, [256, False]],  # 17 (P3/8-small)
+
+   [-1, 1, Conv, [256, 3, 2]],
+   [[-1, 14], 1, Concat, [1]],  # cat head P4
+   [-1, 3, C3, [512, False]],  # 20 (P4/16-medium)
+
+   [-1, 1, Conv, [512, 3, 2]],
+   [[-1, 10], 1, Concat, [1]],  # cat head P5
+   [-1, 3, C3, [1024, False]],  # 23 (P5/32-large)
+
+   [[17, 20, 23], 1, Detect, [nc, anchors]],  # Detect(P3, P4, P5)
+  ]

models/yolov5m.yaml

@@ -0,0 +1,48 @@
+# YOLOv5 🚀 by Ultralytics, AGPL-3.0 license
+
+# Parameters
+nc: 80  # number of classes
+depth_multiple: 0.67  # model depth multiple
+width_multiple: 0.75  # layer channel multiple
+anchors:
+  - [10,13, 16,30, 33,23]  # P3/8
+  - [30,61, 62,45, 59,119]  # P4/16
+  - [116,90, 156,198, 373,326]  # P5/32
+
+# YOLOv5 v6.0 backbone
+backbone:
+  # [from, number, module, args]
+  [[-1, 1, Conv, [64, 6, 2, 2]],  # 0-P1/2
+   [-1, 1, Conv, [128, 3, 2]],  # 1-P2/4
+   [-1, 3, C3, [128]],
+   [-1, 1, Conv, [256, 3, 2]],  # 3-P3/8
+   [-1, 6, C3, [256]],
+   [-1, 1, Conv, [512, 3, 2]],  # 5-P4/16
+   [-1, 9, C3, [512]],
+   [-1, 1, Conv, [1024, 3, 2]],  # 7-P5/32
+   [-1, 3, C3, [1024]],
+   [-1, 1, SPPF, [1024, 5]],  # 9
+  ]
+
+# YOLOv5 v6.0 head
+head:
+  [[-1, 1, Conv, [512, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 6], 1, Concat, [1]],  # cat backbone P4
+   [-1, 3, C3, [512, False]],  # 13
+
+   [-1, 1, Conv, [256, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 4], 1, Concat, [1]],  # cat backbone P3
+   [-1, 3, C3, [256, False]],  # 17 (P3/8-small)
+
+   [-1, 1, Conv, [256, 3, 2]],
+   [[-1, 14], 1, Concat, [1]],  # cat head P4
+   [-1, 3, C3, [512, False]],  # 20 (P4/16-medium)
+
+   [-1, 1, Conv, [512, 3, 2]],
+   [[-1, 10], 1, Concat, [1]],  # cat head P5
+   [-1, 3, C3, [1024, False]],  # 23 (P5/32-large)
+
+   [[17, 20, 23], 1, Detect, [nc, anchors]],  # Detect(P3, P4, P5)
+  ]

models/yolov5n.yaml

@@ -0,0 +1,48 @@
+# YOLOv5 🚀 by Ultralytics, AGPL-3.0 license
+
+# Parameters
+nc: 80  # number of classes
+depth_multiple: 0.33  # model depth multiple
+width_multiple: 0.25  # layer channel multiple
+anchors:
+  - [10,13, 16,30, 33,23]  # P3/8
+  - [30,61, 62,45, 59,119]  # P4/16
+  - [116,90, 156,198, 373,326]  # P5/32
+
+# YOLOv5 v6.0 backbone
+backbone:
+  # [from, number, module, args]
+  [[-1, 1, Conv, [64, 6, 2, 2]],  # 0-P1/2
+   [-1, 1, Conv, [128, 3, 2]],  # 1-P2/4
+   [-1, 3, C3, [128]],
+   [-1, 1, Conv, [256, 3, 2]],  # 3-P3/8
+   [-1, 6, C3, [256]],
+   [-1, 1, Conv, [512, 3, 2]],  # 5-P4/16
+   [-1, 9, C3, [512]],
+   [-1, 1, Conv, [1024, 3, 2]],  # 7-P5/32
+   [-1, 3, C3, [1024]],
+   [-1, 1, SPPF, [1024, 5]],  # 9
+  ]
+
+# YOLOv5 v6.0 head
+head:
+  [[-1, 1, Conv, [512, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 6], 1, Concat, [1]],  # cat backbone P4
+   [-1, 3, C3, [512, False]],  # 13
+
+   [-1, 1, Conv, [256, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 4], 1, Concat, [1]],  # cat backbone P3
+   [-1, 3, C3, [256, False]],  # 17 (P3/8-small)
+
+   [-1, 1, Conv, [256, 3, 2]],
+   [[-1, 14], 1, Concat, [1]],  # cat head P4
+   [-1, 3, C3, [512, False]],  # 20 (P4/16-medium)
+
+   [-1, 1, Conv, [512, 3, 2]],
+   [[-1, 10], 1, Concat, [1]],  # cat head P5
+   [-1, 3, C3, [1024, False]],  # 23 (P5/32-large)
+
+   [[17, 20, 23], 1, Detect, [nc, anchors]],  # Detect(P3, P4, P5)
+  ]

models/yolov5s.yaml

@@ -0,0 +1,48 @@
+# YOLOv5 🚀 by Ultralytics, AGPL-3.0 license
+
+# Parameters
+nc: 80  # number of classes
+depth_multiple: 0.33  # model depth multiple
+width_multiple: 0.50  # layer channel multiple
+anchors:
+  - [10,13, 16,30, 33,23]  # P3/8
+  - [30,61, 62,45, 59,119]  # P4/16
+  - [116,90, 156,198, 373,326]  # P5/32
+
+# YOLOv5 v6.0 backbone
+backbone:
+  # [from, number, module, args]
+  [[-1, 1, Conv, [64, 6, 2, 2]],  # 0-P1/2
+   [-1, 1, Conv, [128, 3, 2]],  # 1-P2/4
+   [-1, 3, C3, [128]],
+   [-1, 1, Conv, [256, 3, 2]],  # 3-P3/8
+   [-1, 6, C3, [256]],
+   [-1, 1, Conv, [512, 3, 2]],  # 5-P4/16
+   [-1, 9, C3, [512]],
+   [-1, 1, Conv, [1024, 3, 2]],  # 7-P5/32
+   [-1, 3, C3, [1024]],
+   [-1, 1, SPPF, [1024, 5]],  # 9
+  ]
+
+# YOLOv5 v6.0 head
+head:
+  [[-1, 1, Conv, [512, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 6], 1, Concat, [1]],  # cat backbone P4
+   [-1, 3, C3, [512, False]],  # 13
+
+   [-1, 1, Conv, [256, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 4], 1, Concat, [1]],  # cat backbone P3
+   [-1, 3, C3, [256, False]],  # 17 (P3/8-small)
+
+   [-1, 1, Conv, [256, 3, 2]],
+   [[-1, 14], 1, Concat, [1]],  # cat head P4
+   [-1, 3, C3, [512, False]],  # 20 (P4/16-medium)
+
+   [-1, 1, Conv, [512, 3, 2]],
+   [[-1, 10], 1, Concat, [1]],  # cat head P5
+   [-1, 3, C3, [1024, False]],  # 23 (P5/32-large)
+
+   [[17, 20, 23], 1, Detect, [nc, anchors]],  # Detect(P3, P4, P5)
+  ]

models/yolov5x.yaml

@@ -0,0 +1,48 @@
+# YOLOv5 🚀 by Ultralytics, AGPL-3.0 license
+
+# Parameters
+nc: 80  # number of classes
+depth_multiple: 1.33  # model depth multiple
+width_multiple: 1.25  # layer channel multiple
+anchors:
+  - [10,13, 16,30, 33,23]  # P3/8
+  - [30,61, 62,45, 59,119]  # P4/16
+  - [116,90, 156,198, 373,326]  # P5/32
+
+# YOLOv5 v6.0 backbone
+backbone:
+  # [from, number, module, args]
+  [[-1, 1, Conv, [64, 6, 2, 2]],  # 0-P1/2
+   [-1, 1, Conv, [128, 3, 2]],  # 1-P2/4
+   [-1, 3, C3, [128]],
+   [-1, 1, Conv, [256, 3, 2]],  # 3-P3/8
+   [-1, 6, C3, [256]],
+   [-1, 1, Conv, [512, 3, 2]],  # 5-P4/16
+   [-1, 9, C3, [512]],
+   [-1, 1, Conv, [1024, 3, 2]],  # 7-P5/32
+   [-1, 3, C3, [1024]],
+   [-1, 1, SPPF, [1024, 5]],  # 9
+  ]
+
+# YOLOv5 v6.0 head
+head:
+  [[-1, 1, Conv, [512, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 6], 1, Concat, [1]],  # cat backbone P4
+   [-1, 3, C3, [512, False]],  # 13
+
+   [-1, 1, Conv, [256, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 4], 1, Concat, [1]],  # cat backbone P3
+   [-1, 3, C3, [256, False]],  # 17 (P3/8-small)
+
+   [-1, 1, Conv, [256, 3, 2]],
+   [[-1, 14], 1, Concat, [1]],  # cat head P4
+   [-1, 3, C3, [512, False]],  # 20 (P4/16-medium)
+
+   [-1, 1, Conv, [512, 3, 2]],
+   [[-1, 10], 1, Concat, [1]],  # cat head P5
+   [-1, 3, C3, [1024, False]],  # 23 (P5/32-large)
+
+   [[17, 20, 23], 1, Detect, [nc, anchors]],  # Detect(P3, P4, P5)
+  ]

requirements.txt

@@ -0,0 +1,51 @@
+# YOLOv5 requirements
+# Usage: pip install -r requirements.txt
+
+gradio
+torch
+# Base ------------------------------------------------------------------------
+gitpython>=3.1.30
+matplotlib>=3.3
+numpy>=1.22.2
+opencv-python>=4.1.1
+Pillow>=7.1.2
+psutil  # system resources
+PyYAML>=5.3.1
+requests>=2.23.0
+scipy>=1.4.1
+thop>=0.1.1  # FLOPs computation
+torch>=1.8.0  # see https://pytorch.org/get-started/locally (recommended)
+torchvision>=0.9.0
+tqdm>=4.64.0
+ultralytics>=8.0.147
+# protobuf<=3.20.1  # https://github.com/ultralytics/yolov5/issues/8012
+
+# Logging ---------------------------------------------------------------------
+# tensorboard>=2.4.1
+# clearml>=1.2.0
+# comet
+
+# Plotting --------------------------------------------------------------------
+pandas>=1.1.4
+seaborn>=0.11.0
+
+# Export ----------------------------------------------------------------------
+# coremltools>=6.0  # CoreML export
+# onnx>=1.10.0  # ONNX export
+# onnx-simplifier>=0.4.1  # ONNX simplifier
+# nvidia-pyindex  # TensorRT export
+# nvidia-tensorrt  # TensorRT export
+# scikit-learn<=1.1.2  # CoreML quantization
+# tensorflow>=2.4.0  # TF exports (-cpu, -aarch64, -macos)
+# tensorflowjs>=3.9.0  # TF.js export
+# openvino-dev>=2023.0  # OpenVINO export
+
+# Deploy ----------------------------------------------------------------------
+setuptools>=65.5.1 # Snyk vulnerability fix
+# tritonclient[all]~=2.24.0
+
+# Extras ----------------------------------------------------------------------
+# ipython  # interactive notebook
+# mss  # screenshots
+# albumentations>=1.0.3
+# pycocotools>=2.0.6  # COCO mAP

utils/__init__.py

@@ -0,0 +1,86 @@
+# YOLOv5 🚀 by Ultralytics, AGPL-3.0 license
+"""
+utils/initialization
+"""
+
+import contextlib
+import platform
+import threading
+
+
+def emojis(str=''):
+    # Return platform-dependent emoji-safe version of string
+    return str.encode().decode('ascii', 'ignore') if platform.system() == 'Windows' else str
+
+
+class TryExcept(contextlib.ContextDecorator):
+    # YOLOv5 TryExcept class. Usage: @TryExcept() decorator or 'with TryExcept():' context manager
+    def __init__(self, msg=''):
+        self.msg = msg
+
+    def __enter__(self):
+        pass
+
+    def __exit__(self, exc_type, value, traceback):
+        if value:
+            print(emojis(f"{self.msg}{': ' if self.msg else ''}{value}"))
+        return True
+
+
+def threaded(func):
+    # Multi-threads a target function and returns thread. Usage: @threaded decorator
+    def wrapper(*args, **kwargs):
+        thread = threading.Thread(target=func, args=args, kwargs=kwargs, daemon=True)
+        thread.start()
+        return thread
+
+    return wrapper
+
+
+def join_threads(verbose=False):
+    # Join all daemon threads, i.e. atexit.register(lambda: join_threads())
+    main_thread = threading.current_thread()
+    for t in threading.enumerate():
+        if t is not main_thread:
+            if verbose:
+                print(f'Joining thread {t.name}')
+            t.join()
+
+
+def notebook_init(verbose=True):
+    # Check system software and hardware
+    print('Checking setup...')
+
+    import os
+    import shutil
+
+    from ultralytics.utils.checks import check_requirements
+
+    from utils.general import check_font, is_colab
+    from utils.torch_utils import select_device  # imports
+
+    check_font()
+
+    import psutil
+
+    if check_requirements('wandb', install=False):
+        os.system('pip uninstall -y wandb')  # eliminate unexpected account creation prompt with infinite hang
+    if is_colab():
+        shutil.rmtree('/content/sample_data', ignore_errors=True)  # remove colab /sample_data directory
+
+    # System info
+    display = None
+    if verbose:
+        gb = 1 << 30  # bytes to GiB (1024 ** 3)
+        ram = psutil.virtual_memory().total
+        total, used, free = shutil.disk_usage('/')
+        with contextlib.suppress(Exception):  # clear display if ipython is installed
+            from IPython import display
+            display.clear_output()
+        s = f'({os.cpu_count()} CPUs, {ram / gb:.1f} GB RAM, {(total - free) / gb:.1f}/{total / gb:.1f} GB disk)'
+    else:
+        s = ''
+
+    select_device(newline=False)
+    print(emojis(f'Setup complete ✅ {s}'))
+    return display

utils/activations.py

@@ -0,0 +1,103 @@
+# YOLOv5 🚀 by Ultralytics, AGPL-3.0 license
+"""
+Activation functions
+"""
+
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+
+class SiLU(nn.Module):
+    # SiLU activation https://arxiv.org/pdf/1606.08415.pdf
+    @staticmethod
+    def forward(x):
+        return x * torch.sigmoid(x)
+
+
+class Hardswish(nn.Module):
+    # Hard-SiLU activation
+    @staticmethod
+    def forward(x):
+        # return x * F.hardsigmoid(x)  # for TorchScript and CoreML
+        return x * F.hardtanh(x + 3, 0.0, 6.0) / 6.0  # for TorchScript, CoreML and ONNX
+
+
+class Mish(nn.Module):
+    # Mish activation https://github.com/digantamisra98/Mish
+    @staticmethod
+    def forward(x):
+        return x * F.softplus(x).tanh()
+
+
+class MemoryEfficientMish(nn.Module):
+    # Mish activation memory-efficient
+    class F(torch.autograd.Function):
+
+        @staticmethod
+        def forward(ctx, x):
+            ctx.save_for_backward(x)
+            return x.mul(torch.tanh(F.softplus(x)))  # x * tanh(ln(1 + exp(x)))
+
+        @staticmethod
+        def backward(ctx, grad_output):
+            x = ctx.saved_tensors[0]
+            sx = torch.sigmoid(x)
+            fx = F.softplus(x).tanh()
+            return grad_output * (fx + x * sx * (1 - fx * fx))
+
+    def forward(self, x):
+        return self.F.apply(x)
+
+
+class FReLU(nn.Module):
+    # FReLU activation https://arxiv.org/abs/2007.11824
+    def __init__(self, c1, k=3):  # ch_in, kernel
+        super().__init__()
+        self.conv = nn.Conv2d(c1, c1, k, 1, 1, groups=c1, bias=False)
+        self.bn = nn.BatchNorm2d(c1)
+
+    def forward(self, x):
+        return torch.max(x, self.bn(self.conv(x)))
+
+
+class AconC(nn.Module):
+    r""" ACON activation (activate or not)
+    AconC: (p1*x-p2*x) * sigmoid(beta*(p1*x-p2*x)) + p2*x, beta is a learnable parameter
+    according to "Activate or Not: Learning Customized Activation" <https://arxiv.org/pdf/2009.04759.pdf>.
+    """
+
+    def __init__(self, c1):
+        super().__init__()
+        self.p1 = nn.Parameter(torch.randn(1, c1, 1, 1))
+        self.p2 = nn.Parameter(torch.randn(1, c1, 1, 1))
+        self.beta = nn.Parameter(torch.ones(1, c1, 1, 1))
+
+    def forward(self, x):
+        dpx = (self.p1 - self.p2) * x
+        return dpx * torch.sigmoid(self.beta * dpx) + self.p2 * x
+
+
+class MetaAconC(nn.Module):
+    r""" ACON activation (activate or not)
+    MetaAconC: (p1*x-p2*x) * sigmoid(beta*(p1*x-p2*x)) + p2*x, beta is generated by a small network
+    according to "Activate or Not: Learning Customized Activation" <https://arxiv.org/pdf/2009.04759.pdf>.
+    """
+
+    def __init__(self, c1, k=1, s=1, r=16):  # ch_in, kernel, stride, r
+        super().__init__()
+        c2 = max(r, c1 // r)
+        self.p1 = nn.Parameter(torch.randn(1, c1, 1, 1))
+        self.p2 = nn.Parameter(torch.randn(1, c1, 1, 1))
+        self.fc1 = nn.Conv2d(c1, c2, k, s, bias=True)
+        self.fc2 = nn.Conv2d(c2, c1, k, s, bias=True)
+        # self.bn1 = nn.BatchNorm2d(c2)
+        # self.bn2 = nn.BatchNorm2d(c1)
+
+    def forward(self, x):
+        y = x.mean(dim=2, keepdims=True).mean(dim=3, keepdims=True)
+        # batch-size 1 bug/instabilities https://github.com/ultralytics/yolov5/issues/2891
+        # beta = torch.sigmoid(self.bn2(self.fc2(self.bn1(self.fc1(y)))))  # bug/unstable
+        beta = torch.sigmoid(self.fc2(self.fc1(y)))  # bug patch BN layers removed
+        dpx = (self.p1 - self.p2) * x
+        return dpx * torch.sigmoid(beta * dpx) + self.p2 * x

utils/augmentations.py

@@ -0,0 +1,397 @@
+# YOLOv5 🚀 by Ultralytics, AGPL-3.0 license
+"""
+Image augmentation functions
+"""
+
+import math
+import random
+
+import cv2
+import numpy as np
+import torch
+import torchvision.transforms as T
+import torchvision.transforms.functional as TF
+
+from utils.general import LOGGER, check_version, colorstr, resample_segments, segment2box, xywhn2xyxy
+from utils.metrics import bbox_ioa
+
+IMAGENET_MEAN = 0.485, 0.456, 0.406  # RGB mean
+IMAGENET_STD = 0.229, 0.224, 0.225  # RGB standard deviation
+
+
+class Albumentations:
+    # YOLOv5 Albumentations class (optional, only used if package is installed)
+    def __init__(self, size=640):
+        self.transform = None
+        prefix = colorstr('albumentations: ')
+        try:
+            import albumentations as A
+            check_version(A.__version__, '1.0.3', hard=True)  # version requirement
+
+            T = [
+                A.RandomResizedCrop(height=size, width=size, scale=(0.8, 1.0), ratio=(0.9, 1.11), p=0.0),
+                A.Blur(p=0.01),
+                A.MedianBlur(p=0.01),
+                A.ToGray(p=0.01),
+                A.CLAHE(p=0.01),
+                A.RandomBrightnessContrast(p=0.0),
+                A.RandomGamma(p=0.0),
+                A.ImageCompression(quality_lower=75, p=0.0)]  # transforms
+            self.transform = A.Compose(T, bbox_params=A.BboxParams(format='yolo', label_fields=['class_labels']))
+
+            LOGGER.info(prefix + ', '.join(f'{x}'.replace('always_apply=False, ', '') for x in T if x.p))
+        except ImportError:  # package not installed, skip
+            pass
+        except Exception as e:
+            LOGGER.info(f'{prefix}{e}')
+
+    def __call__(self, im, labels, p=1.0):
+        if self.transform and random.random() < p:
+            new = self.transform(image=im, bboxes=labels[:, 1:], class_labels=labels[:, 0])  # transformed
+            im, labels = new['image'], np.array([[c, *b] for c, b in zip(new['class_labels'], new['bboxes'])])
+        return im, labels
+
+
+def normalize(x, mean=IMAGENET_MEAN, std=IMAGENET_STD, inplace=False):
+    # Denormalize RGB images x per ImageNet stats in BCHW format, i.e. = (x - mean) / std
+    return TF.normalize(x, mean, std, inplace=inplace)
+
+
+def denormalize(x, mean=IMAGENET_MEAN, std=IMAGENET_STD):
+    # Denormalize RGB images x per ImageNet stats in BCHW format, i.e. = x * std + mean
+    for i in range(3):
+        x[:, i] = x[:, i] * std[i] + mean[i]
+    return x
+
+
+def augment_hsv(im, hgain=0.5, sgain=0.5, vgain=0.5):
+    # HSV color-space augmentation
+    if hgain or sgain or vgain:
+        r = np.random.uniform(-1, 1, 3) * [hgain, sgain, vgain] + 1  # random gains
+        hue, sat, val = cv2.split(cv2.cvtColor(im, cv2.COLOR_BGR2HSV))
+        dtype = im.dtype  # uint8
+
+        x = np.arange(0, 256, dtype=r.dtype)
+        lut_hue = ((x * r[0]) % 180).astype(dtype)
+        lut_sat = np.clip(x * r[1], 0, 255).astype(dtype)
+        lut_val = np.clip(x * r[2], 0, 255).astype(dtype)
+
+        im_hsv = cv2.merge((cv2.LUT(hue, lut_hue), cv2.LUT(sat, lut_sat), cv2.LUT(val, lut_val)))
+        cv2.cvtColor(im_hsv, cv2.COLOR_HSV2BGR, dst=im)  # no return needed
+
+
+def hist_equalize(im, clahe=True, bgr=False):
+    # Equalize histogram on BGR image 'im' with im.shape(n,m,3) and range 0-255
+    yuv = cv2.cvtColor(im, cv2.COLOR_BGR2YUV if bgr else cv2.COLOR_RGB2YUV)
+    if clahe:
+        c = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8, 8))
+        yuv[:, :, 0] = c.apply(yuv[:, :, 0])
+    else:
+        yuv[:, :, 0] = cv2.equalizeHist(yuv[:, :, 0])  # equalize Y channel histogram
+    return cv2.cvtColor(yuv, cv2.COLOR_YUV2BGR if bgr else cv2.COLOR_YUV2RGB)  # convert YUV image to RGB
+
+
+def replicate(im, labels):
+    # Replicate labels
+    h, w = im.shape[:2]
+    boxes = labels[:, 1:].astype(int)
+    x1, y1, x2, y2 = boxes.T
+    s = ((x2 - x1) + (y2 - y1)) / 2  # side length (pixels)
+    for i in s.argsort()[:round(s.size * 0.5)]:  # smallest indices
+        x1b, y1b, x2b, y2b = boxes[i]
+        bh, bw = y2b - y1b, x2b - x1b
+        yc, xc = int(random.uniform(0, h - bh)), int(random.uniform(0, w - bw))  # offset x, y
+        x1a, y1a, x2a, y2a = [xc, yc, xc + bw, yc + bh]
+        im[y1a:y2a, x1a:x2a] = im[y1b:y2b, x1b:x2b]  # im4[ymin:ymax, xmin:xmax]
+        labels = np.append(labels, [[labels[i, 0], x1a, y1a, x2a, y2a]], axis=0)
+
+    return im, labels
+
+
+def letterbox(im, new_shape=(640, 640), color=(114, 114, 114), auto=True, scaleFill=False, scaleup=True, stride=32):
+    # Resize and pad image while meeting stride-multiple constraints
+    shape = im.shape[:2]  # current shape [height, width]
+    if isinstance(new_shape, int):
+        new_shape = (new_shape, new_shape)
+
+    # Scale ratio (new / old)
+    r = min(new_shape[0] / shape[0], new_shape[1] / shape[1])
+    if not scaleup:  # only scale down, do not scale up (for better val mAP)
+        r = min(r, 1.0)
+
+    # Compute padding
+    ratio = r, r  # width, height ratios
+    new_unpad = int(round(shape[1] * r)), int(round(shape[0] * r))
+    dw, dh = new_shape[1] - new_unpad[0], new_shape[0] - new_unpad[1]  # wh padding
+    if auto:  # minimum rectangle
+        dw, dh = np.mod(dw, stride), np.mod(dh, stride)  # wh padding
+    elif scaleFill:  # stretch
+        dw, dh = 0.0, 0.0
+        new_unpad = (new_shape[1], new_shape[0])
+        ratio = new_shape[1] / shape[1], new_shape[0] / shape[0]  # width, height ratios
+
+    dw /= 2  # divide padding into 2 sides
+    dh /= 2
+
+    if shape[::-1] != new_unpad:  # resize
+        im = cv2.resize(im, new_unpad, interpolation=cv2.INTER_LINEAR)
+    top, bottom = int(round(dh - 0.1)), int(round(dh + 0.1))
+    left, right = int(round(dw - 0.1)), int(round(dw + 0.1))
+    im = cv2.copyMakeBorder(im, top, bottom, left, right, cv2.BORDER_CONSTANT, value=color)  # add border
+    return im, ratio, (dw, dh)
+
+
+def random_perspective(im,
+                       targets=(),
+                       segments=(),
+                       degrees=10,
+                       translate=.1,
+                       scale=.1,
+                       shear=10,
+                       perspective=0.0,
+                       border=(0, 0)):
+    # torchvision.transforms.RandomAffine(degrees=(-10, 10), translate=(0.1, 0.1), scale=(0.9, 1.1), shear=(-10, 10))
+    # targets = [cls, xyxy]
+
+    height = im.shape[0] + border[0] * 2  # shape(h,w,c)
+    width = im.shape[1] + border[1] * 2
+
+    # Center
+    C = np.eye(3)
+    C[0, 2] = -im.shape[1] / 2  # x translation (pixels)
+    C[1, 2] = -im.shape[0] / 2  # y translation (pixels)
+
+    # Perspective
+    P = np.eye(3)
+    P[2, 0] = random.uniform(-perspective, perspective)  # x perspective (about y)
+    P[2, 1] = random.uniform(-perspective, perspective)  # y perspective (about x)
+
+    # Rotation and Scale
+    R = np.eye(3)
+    a = random.uniform(-degrees, degrees)
+    # a += random.choice([-180, -90, 0, 90])  # add 90deg rotations to small rotations
+    s = random.uniform(1 - scale, 1 + scale)
+    # s = 2 ** random.uniform(-scale, scale)
+    R[:2] = cv2.getRotationMatrix2D(angle=a, center=(0, 0), scale=s)
+
+    # Shear
+    S = np.eye(3)
+    S[0, 1] = math.tan(random.uniform(-shear, shear) * math.pi / 180)  # x shear (deg)
+    S[1, 0] = math.tan(random.uniform(-shear, shear) * math.pi / 180)  # y shear (deg)
+
+    # Translation
+    T = np.eye(3)
+    T[0, 2] = random.uniform(0.5 - translate, 0.5 + translate) * width  # x translation (pixels)
+    T[1, 2] = random.uniform(0.5 - translate, 0.5 + translate) * height  # y translation (pixels)
+
+    # Combined rotation matrix
+    M = T @ S @ R @ P @ C  # order of operations (right to left) is IMPORTANT
+    if (border[0] != 0) or (border[1] != 0) or (M != np.eye(3)).any():  # image changed
+        if perspective:
+            im = cv2.warpPerspective(im, M, dsize=(width, height), borderValue=(114, 114, 114))
+        else:  # affine
+            im = cv2.warpAffine(im, M[:2], dsize=(width, height), borderValue=(114, 114, 114))
+
+    # Visualize
+    # import matplotlib.pyplot as plt
+    # ax = plt.subplots(1, 2, figsize=(12, 6))[1].ravel()
+    # ax[0].imshow(im[:, :, ::-1])  # base
+    # ax[1].imshow(im2[:, :, ::-1])  # warped
+
+    # Transform label coordinates
+    n = len(targets)
+    if n:
+        use_segments = any(x.any() for x in segments) and len(segments) == n
+        new = np.zeros((n, 4))
+        if use_segments:  # warp segments
+            segments = resample_segments(segments)  # upsample
+            for i, segment in enumerate(segments):
+                xy = np.ones((len(segment), 3))
+                xy[:, :2] = segment
+                xy = xy @ M.T  # transform
+                xy = xy[:, :2] / xy[:, 2:3] if perspective else xy[:, :2]  # perspective rescale or affine
+
+                # clip
+                new[i] = segment2box(xy, width, height)
+
+        else:  # warp boxes
+            xy = np.ones((n * 4, 3))
+            xy[:, :2] = targets[:, [1, 2, 3, 4, 1, 4, 3, 2]].reshape(n * 4, 2)  # x1y1, x2y2, x1y2, x2y1
+            xy = xy @ M.T  # transform
+            xy = (xy[:, :2] / xy[:, 2:3] if perspective else xy[:, :2]).reshape(n, 8)  # perspective rescale or affine
+
+            # create new boxes
+            x = xy[:, [0, 2, 4, 6]]
+            y = xy[:, [1, 3, 5, 7]]
+            new = np.concatenate((x.min(1), y.min(1), x.max(1), y.max(1))).reshape(4, n).T
+
+            # clip
+            new[:, [0, 2]] = new[:, [0, 2]].clip(0, width)
+            new[:, [1, 3]] = new[:, [1, 3]].clip(0, height)
+
+        # filter candidates
+        i = box_candidates(box1=targets[:, 1:5].T * s, box2=new.T, area_thr=0.01 if use_segments else 0.10)
+        targets = targets[i]
+        targets[:, 1:5] = new[i]
+
+    return im, targets
+
+
+def copy_paste(im, labels, segments, p=0.5):
+    # Implement Copy-Paste augmentation https://arxiv.org/abs/2012.07177, labels as nx5 np.array(cls, xyxy)
+    n = len(segments)
+    if p and n:
+        h, w, c = im.shape  # height, width, channels
+        im_new = np.zeros(im.shape, np.uint8)
+        for j in random.sample(range(n), k=round(p * n)):
+            l, s = labels[j], segments[j]
+            box = w - l[3], l[2], w - l[1], l[4]
+            ioa = bbox_ioa(box, labels[:, 1:5])  # intersection over area
+            if (ioa < 0.30).all():  # allow 30% obscuration of existing labels
+                labels = np.concatenate((labels, [[l[0], *box]]), 0)
+                segments.append(np.concatenate((w - s[:, 0:1], s[:, 1:2]), 1))
+                cv2.drawContours(im_new, [segments[j].astype(np.int32)], -1, (1, 1, 1), cv2.FILLED)
+
+        result = cv2.flip(im, 1)  # augment segments (flip left-right)
+        i = cv2.flip(im_new, 1).astype(bool)
+        im[i] = result[i]  # cv2.imwrite('debug.jpg', im)  # debug
+
+    return im, labels, segments
+
+
+def cutout(im, labels, p=0.5):
+    # Applies image cutout augmentation https://arxiv.org/abs/1708.04552
+    if random.random() < p:
+        h, w = im.shape[:2]
+        scales = [0.5] * 1 + [0.25] * 2 + [0.125] * 4 + [0.0625] * 8 + [0.03125] * 16  # image size fraction
+        for s in scales:
+            mask_h = random.randint(1, int(h * s))  # create random masks
+            mask_w = random.randint(1, int(w * s))
+
+            # box
+            xmin = max(0, random.randint(0, w) - mask_w // 2)
+            ymin = max(0, random.randint(0, h) - mask_h // 2)
+            xmax = min(w, xmin + mask_w)
+            ymax = min(h, ymin + mask_h)
+
+            # apply random color mask
+            im[ymin:ymax, xmin:xmax] = [random.randint(64, 191) for _ in range(3)]
+
+            # return unobscured labels
+            if len(labels) and s > 0.03:
+                box = np.array([xmin, ymin, xmax, ymax], dtype=np.float32)
+                ioa = bbox_ioa(box, xywhn2xyxy(labels[:, 1:5], w, h))  # intersection over area
+                labels = labels[ioa < 0.60]  # remove >60% obscured labels
+
+    return labels
+
+
+def mixup(im, labels, im2, labels2):
+    # Applies MixUp augmentation https://arxiv.org/pdf/1710.09412.pdf
+    r = np.random.beta(32.0, 32.0)  # mixup ratio, alpha=beta=32.0
+    im = (im * r + im2 * (1 - r)).astype(np.uint8)
+    labels = np.concatenate((labels, labels2), 0)
+    return im, labels
+
+
+def box_candidates(box1, box2, wh_thr=2, ar_thr=100, area_thr=0.1, eps=1e-16):  # box1(4,n), box2(4,n)
+    # Compute candidate boxes: box1 before augment, box2 after augment, wh_thr (pixels), aspect_ratio_thr, area_ratio
+    w1, h1 = box1[2] - box1[0], box1[3] - box1[1]
+    w2, h2 = box2[2] - box2[0], box2[3] - box2[1]
+    ar = np.maximum(w2 / (h2 + eps), h2 / (w2 + eps))  # aspect ratio
+    return (w2 > wh_thr) & (h2 > wh_thr) & (w2 * h2 / (w1 * h1 + eps) > area_thr) & (ar < ar_thr)  # candidates
+
+
+def classify_albumentations(
+        augment=True,
+        size=224,
+        scale=(0.08, 1.0),
+        ratio=(0.75, 1.0 / 0.75),  # 0.75, 1.33
+        hflip=0.5,
+        vflip=0.0,
+        jitter=0.4,
+        mean=IMAGENET_MEAN,
+        std=IMAGENET_STD,
+        auto_aug=False):
+    # YOLOv5 classification Albumentations (optional, only used if package is installed)
+    prefix = colorstr('albumentations: ')
+    try:
+        import albumentations as A
+        from albumentations.pytorch import ToTensorV2
+        check_version(A.__version__, '1.0.3', hard=True)  # version requirement
+        if augment:  # Resize and crop
+            T = [A.RandomResizedCrop(height=size, width=size, scale=scale, ratio=ratio)]
+            if auto_aug:
+                # TODO: implement AugMix, AutoAug & RandAug in albumentation
+                LOGGER.info(f'{prefix}auto augmentations are currently not supported')
+            else:
+                if hflip > 0:
+                    T += [A.HorizontalFlip(p=hflip)]
+                if vflip > 0:
+                    T += [A.VerticalFlip(p=vflip)]
+                if jitter > 0:
+                    color_jitter = (float(jitter), ) * 3  # repeat value for brightness, contrast, satuaration, 0 hue
+                    T += [A.ColorJitter(*color_jitter, 0)]
+        else:  # Use fixed crop for eval set (reproducibility)
+            T = [A.SmallestMaxSize(max_size=size), A.CenterCrop(height=size, width=size)]
+        T += [A.Normalize(mean=mean, std=std), ToTensorV2()]  # Normalize and convert to Tensor
+        LOGGER.info(prefix + ', '.join(f'{x}'.replace('always_apply=False, ', '') for x in T if x.p))
+        return A.Compose(T)
+
+    except ImportError:  # package not installed, skip
+        LOGGER.warning(f'{prefix}⚠️ not found, install with `pip install albumentations` (recommended)')
+    except Exception as e:
+        LOGGER.info(f'{prefix}{e}')
+
+
+def classify_transforms(size=224):
+    # Transforms to apply if albumentations not installed
+    assert isinstance(size, int), f'ERROR: classify_transforms size {size} must be integer, not (list, tuple)'
+    # T.Compose([T.ToTensor(), T.Resize(size), T.CenterCrop(size), T.Normalize(IMAGENET_MEAN, IMAGENET_STD)])
+    return T.Compose([CenterCrop(size), ToTensor(), T.Normalize(IMAGENET_MEAN, IMAGENET_STD)])
+
+
+class LetterBox:
+    # YOLOv5 LetterBox class for image preprocessing, i.e. T.Compose([LetterBox(size), ToTensor()])
+    def __init__(self, size=(640, 640), auto=False, stride=32):
+        super().__init__()
+        self.h, self.w = (size, size) if isinstance(size, int) else size
+        self.auto = auto  # pass max size integer, automatically solve for short side using stride
+        self.stride = stride  # used with auto
+
+    def __call__(self, im):  # im = np.array HWC
+        imh, imw = im.shape[:2]
+        r = min(self.h / imh, self.w / imw)  # ratio of new/old
+        h, w = round(imh * r), round(imw * r)  # resized image
+        hs, ws = (math.ceil(x / self.stride) * self.stride for x in (h, w)) if self.auto else self.h, self.w
+        top, left = round((hs - h) / 2 - 0.1), round((ws - w) / 2 - 0.1)
+        im_out = np.full((self.h, self.w, 3), 114, dtype=im.dtype)
+        im_out[top:top + h, left:left + w] = cv2.resize(im, (w, h), interpolation=cv2.INTER_LINEAR)
+        return im_out
+
+
+class CenterCrop:
+    # YOLOv5 CenterCrop class for image preprocessing, i.e. T.Compose([CenterCrop(size), ToTensor()])
+    def __init__(self, size=640):
+        super().__init__()
+        self.h, self.w = (size, size) if isinstance(size, int) else size
+
+    def __call__(self, im):  # im = np.array HWC
+        imh, imw = im.shape[:2]
+        m = min(imh, imw)  # min dimension
+        top, left = (imh - m) // 2, (imw - m) // 2
+        return cv2.resize(im[top:top + m, left:left + m], (self.w, self.h), interpolation=cv2.INTER_LINEAR)
+
+
+class ToTensor:
+    # YOLOv5 ToTensor class for image preprocessing, i.e. T.Compose([LetterBox(size), ToTensor()])
+    def __init__(self, half=False):
+        super().__init__()
+        self.half = half
+
+    def __call__(self, im):  # im = np.array HWC in BGR order
+        im = np.ascontiguousarray(im.transpose((2, 0, 1))[::-1])  # HWC to CHW -> BGR to RGB -> contiguous
+        im = torch.from_numpy(im)  # to torch
+        im = im.half() if self.half else im.float()  # uint8 to fp16/32
+        im /= 255.0  # 0-255 to 0.0-1.0
+        return im

utils/autoanchor.py

@@ -0,0 +1,169 @@
+# YOLOv5 🚀 by Ultralytics, AGPL-3.0 license
+"""
+AutoAnchor utils
+"""
+
+import random
+
+import numpy as np
+import torch
+import yaml
+from tqdm import tqdm
+
+from utils import TryExcept
+from utils.general import LOGGER, TQDM_BAR_FORMAT, colorstr
+
+PREFIX = colorstr('AutoAnchor: ')
+
+
+def check_anchor_order(m):
+    # Check anchor order against stride order for YOLOv5 Detect() module m, and correct if necessary
+    a = m.anchors.prod(-1).mean(-1).view(-1)  # mean anchor area per output layer
+    da = a[-1] - a[0]  # delta a
+    ds = m.stride[-1] - m.stride[0]  # delta s
+    if da and (da.sign() != ds.sign()):  # same order
+        LOGGER.info(f'{PREFIX}Reversing anchor order')
+        m.anchors[:] = m.anchors.flip(0)
+
+
+@TryExcept(f'{PREFIX}ERROR')
+def check_anchors(dataset, model, thr=4.0, imgsz=640):
+    # Check anchor fit to data, recompute if necessary
+    m = model.module.model[-1] if hasattr(model, 'module') else model.model[-1]  # Detect()
+    shapes = imgsz * dataset.shapes / dataset.shapes.max(1, keepdims=True)
+    scale = np.random.uniform(0.9, 1.1, size=(shapes.shape[0], 1))  # augment scale
+    wh = torch.tensor(np.concatenate([l[:, 3:5] * s for s, l in zip(shapes * scale, dataset.labels)])).float()  # wh
+
+    def metric(k):  # compute metric
+        r = wh[:, None] / k[None]
+        x = torch.min(r, 1 / r).min(2)[0]  # ratio metric
+        best = x.max(1)[0]  # best_x
+        aat = (x > 1 / thr).float().sum(1).mean()  # anchors above threshold
+        bpr = (best > 1 / thr).float().mean()  # best possible recall
+        return bpr, aat
+
+    stride = m.stride.to(m.anchors.device).view(-1, 1, 1)  # model strides
+    anchors = m.anchors.clone() * stride  # current anchors
+    bpr, aat = metric(anchors.cpu().view(-1, 2))
+    s = f'\n{PREFIX}{aat:.2f} anchors/target, {bpr:.3f} Best Possible Recall (BPR). '
+    if bpr > 0.98:  # threshold to recompute
+        LOGGER.info(f'{s}Current anchors are a good fit to dataset ✅')
+    else:
+        LOGGER.info(f'{s}Anchors are a poor fit to dataset ⚠️, attempting to improve...')
+        na = m.anchors.numel() // 2  # number of anchors
+        anchors = kmean_anchors(dataset, n=na, img_size=imgsz, thr=thr, gen=1000, verbose=False)
+        new_bpr = metric(anchors)[0]
+        if new_bpr > bpr:  # replace anchors
+            anchors = torch.tensor(anchors, device=m.anchors.device).type_as(m.anchors)
+            m.anchors[:] = anchors.clone().view_as(m.anchors)
+            check_anchor_order(m)  # must be in pixel-space (not grid-space)
+            m.anchors /= stride
+            s = f'{PREFIX}Done ✅ (optional: update model *.yaml to use these anchors in the future)'
+        else:
+            s = f'{PREFIX}Done ⚠️ (original anchors better than new anchors, proceeding with original anchors)'
+        LOGGER.info(s)
+
+
+def kmean_anchors(dataset='./data/coco128.yaml', n=9, img_size=640, thr=4.0, gen=1000, verbose=True):
+    """ Creates kmeans-evolved anchors from training dataset
+
+        Arguments:
+            dataset: path to data.yaml, or a loaded dataset
+            n: number of anchors
+            img_size: image size used for training
+            thr: anchor-label wh ratio threshold hyperparameter hyp['anchor_t'] used for training, default=4.0
+            gen: generations to evolve anchors using genetic algorithm
+            verbose: print all results
+
+        Return:
+            k: kmeans evolved anchors
+
+        Usage:
+            from utils.autoanchor import *; _ = kmean_anchors()
+    """
+    from scipy.cluster.vq import kmeans
+
+    npr = np.random
+    thr = 1 / thr
+
+    def metric(k, wh):  # compute metrics
+        r = wh[:, None] / k[None]
+        x = torch.min(r, 1 / r).min(2)[0]  # ratio metric
+        # x = wh_iou(wh, torch.tensor(k))  # iou metric
+        return x, x.max(1)[0]  # x, best_x
+
+    def anchor_fitness(k):  # mutation fitness
+        _, best = metric(torch.tensor(k, dtype=torch.float32), wh)
+        return (best * (best > thr).float()).mean()  # fitness
+
+    def print_results(k, verbose=True):
+        k = k[np.argsort(k.prod(1))]  # sort small to large
+        x, best = metric(k, wh0)
+        bpr, aat = (best > thr).float().mean(), (x > thr).float().mean() * n  # best possible recall, anch > thr
+        s = f'{PREFIX}thr={thr:.2f}: {bpr:.4f} best possible recall, {aat:.2f} anchors past thr\n' \
+            f'{PREFIX}n={n}, img_size={img_size}, metric_all={x.mean():.3f}/{best.mean():.3f}-mean/best, ' \
+            f'past_thr={x[x > thr].mean():.3f}-mean: '
+        for x in k:
+            s += '%i,%i, ' % (round(x[0]), round(x[1]))
+        if verbose:
+            LOGGER.info(s[:-2])
+        return k
+
+    if isinstance(dataset, str):  # *.yaml file
+        with open(dataset, errors='ignore') as f:
+            data_dict = yaml.safe_load(f)  # model dict
+        from utils.dataloaders import LoadImagesAndLabels
+        dataset = LoadImagesAndLabels(data_dict['train'], augment=True, rect=True)
+
+    # Get label wh
+    shapes = img_size * dataset.shapes / dataset.shapes.max(1, keepdims=True)
+    wh0 = np.concatenate([l[:, 3:5] * s for s, l in zip(shapes, dataset.labels)])  # wh
+
+    # Filter
+    i = (wh0 < 3.0).any(1).sum()
+    if i:
+        LOGGER.info(f'{PREFIX}WARNING ⚠️ Extremely small objects found: {i} of {len(wh0)} labels are <3 pixels in size')
+    wh = wh0[(wh0 >= 2.0).any(1)].astype(np.float32)  # filter > 2 pixels
+    # wh = wh * (npr.rand(wh.shape[0], 1) * 0.9 + 0.1)  # multiply by random scale 0-1
+
+    # Kmeans init
+    try:
+        LOGGER.info(f'{PREFIX}Running kmeans for {n} anchors on {len(wh)} points...')
+        assert n <= len(wh)  # apply overdetermined constraint
+        s = wh.std(0)  # sigmas for whitening
+        k = kmeans(wh / s, n, iter=30)[0] * s  # points
+        assert n == len(k)  # kmeans may return fewer points than requested if wh is insufficient or too similar
+    except Exception:
+        LOGGER.warning(f'{PREFIX}WARNING ⚠️ switching strategies from kmeans to random init')
+        k = np.sort(npr.rand(n * 2)).reshape(n, 2) * img_size  # random init
+    wh, wh0 = (torch.tensor(x, dtype=torch.float32) for x in (wh, wh0))
+    k = print_results(k, verbose=False)
+
+    # Plot
+    # k, d = [None] * 20, [None] * 20
+    # for i in tqdm(range(1, 21)):
+    #     k[i-1], d[i-1] = kmeans(wh / s, i)  # points, mean distance
+    # fig, ax = plt.subplots(1, 2, figsize=(14, 7), tight_layout=True)
+    # ax = ax.ravel()
+    # ax[0].plot(np.arange(1, 21), np.array(d) ** 2, marker='.')
+    # fig, ax = plt.subplots(1, 2, figsize=(14, 7))  # plot wh
+    # ax[0].hist(wh[wh[:, 0]<100, 0],400)
+    # ax[1].hist(wh[wh[:, 1]<100, 1],400)
+    # fig.savefig('wh.png', dpi=200)
+
+    # Evolve
+    f, sh, mp, s = anchor_fitness(k), k.shape, 0.9, 0.1  # fitness, generations, mutation prob, sigma
+    pbar = tqdm(range(gen), bar_format=TQDM_BAR_FORMAT)  # progress bar
+    for _ in pbar:
+        v = np.ones(sh)
+        while (v == 1).all():  # mutate until a change occurs (prevent duplicates)
+            v = ((npr.random(sh) < mp) * random.random() * npr.randn(*sh) * s + 1).clip(0.3, 3.0)
+        kg = (k.copy() * v).clip(min=2.0)
+        fg = anchor_fitness(kg)
+        if fg > f:
+            f, k = fg, kg.copy()
+            pbar.desc = f'{PREFIX}Evolving anchors with Genetic Algorithm: fitness = {f:.4f}'
+            if verbose:
+                print_results(k, verbose)
+
+    return print_results(k).astype(np.float32)

utils/autobatch.py

@@ -0,0 +1,72 @@
+# YOLOv5 🚀 by Ultralytics, AGPL-3.0 license
+"""
+Auto-batch utils
+"""
+
+from copy import deepcopy
+
+import numpy as np
+import torch
+
+from utils.general import LOGGER, colorstr
+from utils.torch_utils import profile
+
+
+def check_train_batch_size(model, imgsz=640, amp=True):
+    # Check YOLOv5 training batch size
+    with torch.cuda.amp.autocast(amp):
+        return autobatch(deepcopy(model).train(), imgsz)  # compute optimal batch size
+
+
+def autobatch(model, imgsz=640, fraction=0.8, batch_size=16):
+    # Automatically estimate best YOLOv5 batch size to use `fraction` of available CUDA memory
+    # Usage:
+    #     import torch
+    #     from utils.autobatch import autobatch
+    #     model = torch.hub.load('ultralytics/yolov5', 'yolov5s', autoshape=False)
+    #     print(autobatch(model))
+
+    # Check device
+    prefix = colorstr('AutoBatch: ')
+    LOGGER.info(f'{prefix}Computing optimal batch size for --imgsz {imgsz}')
+    device = next(model.parameters()).device  # get model device
+    if device.type == 'cpu':
+        LOGGER.info(f'{prefix}CUDA not detected, using default CPU batch-size {batch_size}')
+        return batch_size
+    if torch.backends.cudnn.benchmark:
+        LOGGER.info(f'{prefix} ⚠️ Requires torch.backends.cudnn.benchmark=False, using default batch-size {batch_size}')
+        return batch_size
+
+    # Inspect CUDA memory
+    gb = 1 << 30  # bytes to GiB (1024 ** 3)
+    d = str(device).upper()  # 'CUDA:0'
+    properties = torch.cuda.get_device_properties(device)  # device properties
+    t = properties.total_memory / gb  # GiB total
+    r = torch.cuda.memory_reserved(device) / gb  # GiB reserved
+    a = torch.cuda.memory_allocated(device) / gb  # GiB allocated
+    f = t - (r + a)  # GiB free
+    LOGGER.info(f'{prefix}{d} ({properties.name}) {t:.2f}G total, {r:.2f}G reserved, {a:.2f}G allocated, {f:.2f}G free')
+
+    # Profile batch sizes
+    batch_sizes = [1, 2, 4, 8, 16]
+    try:
+        img = [torch.empty(b, 3, imgsz, imgsz) for b in batch_sizes]
+        results = profile(img, model, n=3, device=device)
+    except Exception as e:
+        LOGGER.warning(f'{prefix}{e}')
+
+    # Fit a solution
+    y = [x[2] for x in results if x]  # memory [2]
+    p = np.polyfit(batch_sizes[:len(y)], y, deg=1)  # first degree polynomial fit
+    b = int((f * fraction - p[1]) / p[0])  # y intercept (optimal batch size)
+    if None in results:  # some sizes failed
+        i = results.index(None)  # first fail index
+        if b >= batch_sizes[i]:  # y intercept above failure point
+            b = batch_sizes[max(i - 1, 0)]  # select prior safe point
+    if b < 1 or b > 1024:  # b outside of safe range
+        b = batch_size
+        LOGGER.warning(f'{prefix}WARNING ⚠️ CUDA anomaly detected, recommend restart environment and retry command.')
+
+    fraction = (np.polyval(p, b) + r + a) / t  # actual fraction predicted
+    LOGGER.info(f'{prefix}Using batch-size {b} for {d} {t * fraction:.2f}G/{t:.2f}G ({fraction * 100:.0f}%) ✅')
+    return b

utils/aws/mime.sh

@@ -0,0 +1,26 @@
+# AWS EC2 instance startup 'MIME' script https://aws.amazon.com/premiumsupport/knowledge-center/execute-user-data-ec2/
+# This script will run on every instance restart, not only on first start
+# --- DO NOT COPY ABOVE COMMENTS WHEN PASTING INTO USERDATA ---
+
+Content-Type: multipart/mixed; boundary="//"
+MIME-Version: 1.0
+
+--//
+Content-Type: text/cloud-config; charset="us-ascii"
+MIME-Version: 1.0
+Content-Transfer-Encoding: 7bit
+Content-Disposition: attachment; filename="cloud-config.txt"
+
+#cloud-config
+cloud_final_modules:
+- [scripts-user, always]
+
+--//
+Content-Type: text/x-shellscript; charset="us-ascii"
+MIME-Version: 1.0
+Content-Transfer-Encoding: 7bit
+Content-Disposition: attachment; filename="userdata.txt"
+
+#!/bin/bash
+# --- paste contents of userdata.sh here ---
+--//

utils/aws/resume.py

@@ -0,0 +1,40 @@
+# Resume all interrupted trainings in yolov5/ dir including DDP trainings
+# Usage: $ python utils/aws/resume.py
+
+import os
+import sys
+from pathlib import Path
+
+import torch
+import yaml
+
+FILE = Path(__file__).resolve()
+ROOT = FILE.parents[2]  # YOLOv5 root directory
+if str(ROOT) not in sys.path:
+    sys.path.append(str(ROOT))  # add ROOT to PATH
+
+port = 0  # --master_port
+path = Path('').resolve()
+for last in path.rglob('*/**/last.pt'):
+    ckpt = torch.load(last)
+    if ckpt['optimizer'] is None:
+        continue
+
+    # Load opt.yaml
+    with open(last.parent.parent / 'opt.yaml', errors='ignore') as f:
+        opt = yaml.safe_load(f)
+
+    # Get device count
+    d = opt['device'].split(',')  # devices
+    nd = len(d)  # number of devices
+    ddp = nd > 1 or (nd == 0 and torch.cuda.device_count() > 1)  # distributed data parallel
+
+    if ddp:  # multi-GPU
+        port += 1
+        cmd = f'python -m torch.distributed.run --nproc_per_node {nd} --master_port {port} train.py --resume {last}'
+    else:  # single-GPU
+        cmd = f'python train.py --resume {last}'
+
+    cmd += ' > /dev/null 2>&1 &'  # redirect output to dev/null and run in daemon thread
+    print(cmd)
+    os.system(cmd)

utils/aws/userdata.sh

@@ -0,0 +1,27 @@
+#!/bin/bash
+# AWS EC2 instance startup script https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/user-data.html
+# This script will run only once on first instance start (for a re-start script see mime.sh)
+# /home/ubuntu (ubuntu) or /home/ec2-user (amazon-linux) is working dir
+# Use >300 GB SSD
+
+cd home/ubuntu
+if [ ! -d yolov5 ]; then
+  echo "Running first-time script." # install dependencies, download COCO, pull Docker
+  git clone https://github.com/ultralytics/yolov5 -b master && sudo chmod -R 777 yolov5
+  cd yolov5
+  bash data/scripts/get_coco.sh && echo "COCO done." &
+  sudo docker pull ultralytics/yolov5:latest && echo "Docker done." &
+  python -m pip install --upgrade pip && pip install -r requirements.txt && python detect.py && echo "Requirements done." &
+  wait && echo "All tasks done." # finish background tasks
+else
+  echo "Running re-start script." # resume interrupted runs
+  i=0
+  list=$(sudo docker ps -qa) # container list i.e. $'one\ntwo\nthree\nfour'
+  while IFS= read -r id; do
+    ((i++))
+    echo "restarting container $i: $id"
+    sudo docker start $id
+    # sudo docker exec -it $id python train.py --resume # single-GPU
+    sudo docker exec -d $id python utils/aws/resume.py # multi-scenario
+  done <<<"$list"
+fi

utils/callbacks.py

@@ -0,0 +1,76 @@
+# YOLOv5 🚀 by Ultralytics, AGPL-3.0 license
+"""
+Callback utils
+"""
+
+import threading
+
+
+class Callbacks:
+    """"
+    Handles all registered callbacks for YOLOv5 Hooks
+    """
+
+    def __init__(self):
+        # Define the available callbacks
+        self._callbacks = {
+            'on_pretrain_routine_start': [],
+            'on_pretrain_routine_end': [],
+            'on_train_start': [],
+            'on_train_epoch_start': [],
+            'on_train_batch_start': [],
+            'optimizer_step': [],
+            'on_before_zero_grad': [],
+            'on_train_batch_end': [],
+            'on_train_epoch_end': [],
+            'on_val_start': [],
+            'on_val_batch_start': [],
+            'on_val_image_end': [],
+            'on_val_batch_end': [],
+            'on_val_end': [],
+            'on_fit_epoch_end': [],  # fit = train + val
+            'on_model_save': [],
+            'on_train_end': [],
+            'on_params_update': [],
+            'teardown': [], }
+        self.stop_training = False  # set True to interrupt training
+
+    def register_action(self, hook, name='', callback=None):
+        """
+        Register a new action to a callback hook
+
+        Args:
+            hook: The callback hook name to register the action to
+            name: The name of the action for later reference
+            callback: The callback to fire
+        """
+        assert hook in self._callbacks, f"hook '{hook}' not found in callbacks {self._callbacks}"
+        assert callable(callback), f"callback '{callback}' is not callable"
+        self._callbacks[hook].append({'name': name, 'callback': callback})
+
+    def get_registered_actions(self, hook=None):
+        """"
+        Returns all the registered actions by callback hook
+
+        Args:
+            hook: The name of the hook to check, defaults to all
+        """
+        return self._callbacks[hook] if hook else self._callbacks
+
+    def run(self, hook, *args, thread=False, **kwargs):
+        """
+        Loop through the registered actions and fire all callbacks on main thread
+
+        Args:
+            hook: The name of the hook to check, defaults to all
+            args: Arguments to receive from YOLOv5
+            thread: (boolean) Run callbacks in daemon thread
+            kwargs: Keyword Arguments to receive from YOLOv5
+        """
+
+        assert hook in self._callbacks, f"hook '{hook}' not found in callbacks {self._callbacks}"
+        for logger in self._callbacks[hook]:
+            if thread:
+                threading.Thread(target=logger['callback'], args=args, kwargs=kwargs, daemon=True).start()
+            else:
+                logger['callback'](*args, **kwargs)

utils/dataloaders.py

@@ -0,0 +1,1222 @@
+# YOLOv5 🚀 by Ultralytics, AGPL-3.0 license
+"""
+Dataloaders and dataset utils
+"""
+
+import contextlib
+import glob
+import hashlib
+import json
+import math
+import os
+import random
+import shutil
+import time
+from itertools import repeat
+from multiprocessing.pool import Pool, ThreadPool
+from pathlib import Path
+from threading import Thread
+from urllib.parse import urlparse
+
+import numpy as np
+import psutil
+import torch
+import torch.nn.functional as F
+import torchvision
+import yaml
+from PIL import ExifTags, Image, ImageOps
+from torch.utils.data import DataLoader, Dataset, dataloader, distributed
+from tqdm import tqdm
+
+from utils.augmentations import (Albumentations, augment_hsv, classify_albumentations, classify_transforms, copy_paste,
+                                 letterbox, mixup, random_perspective)
+from utils.general import (DATASETS_DIR, LOGGER, NUM_THREADS, TQDM_BAR_FORMAT, check_dataset, check_requirements,
+                           check_yaml, clean_str, cv2, is_colab, is_kaggle, segments2boxes, unzip_file, xyn2xy,
+                           xywh2xyxy, xywhn2xyxy, xyxy2xywhn)
+from utils.torch_utils import torch_distributed_zero_first
+
+# Parameters
+HELP_URL = 'See https://docs.ultralytics.com/yolov5/tutorials/train_custom_data'
+IMG_FORMATS = 'bmp', 'dng', 'jpeg', 'jpg', 'mpo', 'png', 'tif', 'tiff', 'webp', 'pfm'  # include image suffixes
+VID_FORMATS = 'asf', 'avi', 'gif', 'm4v', 'mkv', 'mov', 'mp4', 'mpeg', 'mpg', 'ts', 'wmv'  # include video suffixes
+LOCAL_RANK = int(os.getenv('LOCAL_RANK', -1))  # https://pytorch.org/docs/stable/elastic/run.html
+RANK = int(os.getenv('RANK', -1))
+PIN_MEMORY = str(os.getenv('PIN_MEMORY', True)).lower() == 'true'  # global pin_memory for dataloaders
+
+# Get orientation exif tag
+for orientation in ExifTags.TAGS.keys():
+    if ExifTags.TAGS[orientation] == 'Orientation':
+        break
+
+
+def get_hash(paths):
+    # Returns a single hash value of a list of paths (files or dirs)
+    size = sum(os.path.getsize(p) for p in paths if os.path.exists(p))  # sizes
+    h = hashlib.sha256(str(size).encode())  # hash sizes
+    h.update(''.join(paths).encode())  # hash paths
+    return h.hexdigest()  # return hash
+
+
+def exif_size(img):
+    # Returns exif-corrected PIL size
+    s = img.size  # (width, height)
+    with contextlib.suppress(Exception):
+        rotation = dict(img._getexif().items())[orientation]
+        if rotation in [6, 8]:  # rotation 270 or 90
+            s = (s[1], s[0])
+    return s
+
+
+def exif_transpose(image):
+    """
+    Transpose a PIL image accordingly if it has an EXIF Orientation tag.
+    Inplace version of https://github.com/python-pillow/Pillow/blob/master/src/PIL/ImageOps.py exif_transpose()
+
+    :param image: The image to transpose.
+    :return: An image.
+    """
+    exif = image.getexif()
+    orientation = exif.get(0x0112, 1)  # default 1
+    if orientation > 1:
+        method = {
+            2: Image.FLIP_LEFT_RIGHT,
+            3: Image.ROTATE_180,
+            4: Image.FLIP_TOP_BOTTOM,
+            5: Image.TRANSPOSE,
+            6: Image.ROTATE_270,
+            7: Image.TRANSVERSE,
+            8: Image.ROTATE_90}.get(orientation)
+        if method is not None:
+            image = image.transpose(method)
+            del exif[0x0112]
+            image.info['exif'] = exif.tobytes()
+    return image
+
+
+def seed_worker(worker_id):
+    # Set dataloader worker seed https://pytorch.org/docs/stable/notes/randomness.html#dataloader
+    worker_seed = torch.initial_seed() % 2 ** 32
+    np.random.seed(worker_seed)
+    random.seed(worker_seed)
+
+
+def create_dataloader(path,
+                      imgsz,
+                      batch_size,
+                      stride,
+                      single_cls=False,
+                      hyp=None,
+                      augment=False,
+                      cache=False,
+                      pad=0.0,
+                      rect=False,
+                      rank=-1,
+                      workers=8,
+                      image_weights=False,
+                      quad=False,
+                      prefix='',
+                      shuffle=False,
+                      seed=0):
+    if rect and shuffle:
+        LOGGER.warning('WARNING ⚠️ --rect is incompatible with DataLoader shuffle, setting shuffle=False')
+        shuffle = False
+    with torch_distributed_zero_first(rank):  # init dataset *.cache only once if DDP
+        dataset = LoadImagesAndLabels(
+            path,
+            imgsz,
+            batch_size,
+            augment=augment,  # augmentation
+            hyp=hyp,  # hyperparameters
+            rect=rect,  # rectangular batches
+            cache_images=cache,
+            single_cls=single_cls,
+            stride=int(stride),
+            pad=pad,
+            image_weights=image_weights,
+            prefix=prefix)
+
+    batch_size = min(batch_size, len(dataset))
+    nd = torch.cuda.device_count()  # number of CUDA devices
+    nw = min([os.cpu_count() // max(nd, 1), batch_size if batch_size > 1 else 0, workers])  # number of workers
+    sampler = None if rank == -1 else distributed.DistributedSampler(dataset, shuffle=shuffle)
+    loader = DataLoader if image_weights else InfiniteDataLoader  # only DataLoader allows for attribute updates
+    generator = torch.Generator()
+    generator.manual_seed(6148914691236517205 + seed + RANK)
+    return loader(dataset,
+                  batch_size=batch_size,
+                  shuffle=shuffle and sampler is None,
+                  num_workers=nw,
+                  sampler=sampler,
+                  pin_memory=PIN_MEMORY,
+                  collate_fn=LoadImagesAndLabels.collate_fn4 if quad else LoadImagesAndLabels.collate_fn,
+                  worker_init_fn=seed_worker,
+                  generator=generator), dataset
+
+
+class InfiniteDataLoader(dataloader.DataLoader):
+    """ Dataloader that reuses workers
+
+    Uses same syntax as vanilla DataLoader
+    """
+
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+        object.__setattr__(self, 'batch_sampler', _RepeatSampler(self.batch_sampler))
+        self.iterator = super().__iter__()
+
+    def __len__(self):
+        return len(self.batch_sampler.sampler)
+
+    def __iter__(self):
+        for _ in range(len(self)):
+            yield next(self.iterator)
+
+
+class _RepeatSampler:
+    """ Sampler that repeats forever
+
+    Args:
+        sampler (Sampler)
+    """
+
+    def __init__(self, sampler):
+        self.sampler = sampler
+
+    def __iter__(self):
+        while True:
+            yield from iter(self.sampler)
+
+
+class LoadScreenshots:
+    # YOLOv5 screenshot dataloader, i.e. `python detect.py --source "screen 0 100 100 512 256"`
+    def __init__(self, source, img_size=640, stride=32, auto=True, transforms=None):
+        # source = [screen_number left top width height] (pixels)
+        check_requirements('mss')
+        import mss
+
+        source, *params = source.split()
+        self.screen, left, top, width, height = 0, None, None, None, None  # default to full screen 0
+        if len(params) == 1:
+            self.screen = int(params[0])
+        elif len(params) == 4:
+            left, top, width, height = (int(x) for x in params)
+        elif len(params) == 5:
+            self.screen, left, top, width, height = (int(x) for x in params)
+        self.img_size = img_size
+        self.stride = stride
+        self.transforms = transforms
+        self.auto = auto
+        self.mode = 'stream'
+        self.frame = 0
+        self.sct = mss.mss()
+
+        # Parse monitor shape
+        monitor = self.sct.monitors[self.screen]
+        self.top = monitor['top'] if top is None else (monitor['top'] + top)
+        self.left = monitor['left'] if left is None else (monitor['left'] + left)
+        self.width = width or monitor['width']
+        self.height = height or monitor['height']
+        self.monitor = {'left': self.left, 'top': self.top, 'width': self.width, 'height': self.height}
+
+    def __iter__(self):
+        return self
+
+    def __next__(self):
+        # mss screen capture: get raw pixels from the screen as np array
+        im0 = np.array(self.sct.grab(self.monitor))[:, :, :3]  # [:, :, :3] BGRA to BGR
+        s = f'screen {self.screen} (LTWH): {self.left},{self.top},{self.width},{self.height}: '
+
+        if self.transforms:
+            im = self.transforms(im0)  # transforms
+        else:
+            im = letterbox(im0, self.img_size, stride=self.stride, auto=self.auto)[0]  # padded resize
+            im = im.transpose((2, 0, 1))[::-1]  # HWC to CHW, BGR to RGB
+            im = np.ascontiguousarray(im)  # contiguous
+        self.frame += 1
+        return str(self.screen), im, im0, None, s  # screen, img, original img, im0s, s
+
+
+class LoadImages:
+    # YOLOv5 image/video dataloader, i.e. `python detect.py --source image.jpg/vid.mp4`
+    def __init__(self, path, img_size=640, stride=32, auto=True, transforms=None, vid_stride=1):
+        if isinstance(path, str) and Path(path).suffix == '.txt':  # *.txt file with img/vid/dir on each line
+            path = Path(path).read_text().rsplit()
+        files = []
+        for p in sorted(path) if isinstance(path, (list, tuple)) else [path]:
+            p = str(Path(p).resolve())
+            if '*' in p:
+                files.extend(sorted(glob.glob(p, recursive=True)))  # glob
+            elif os.path.isdir(p):
+                files.extend(sorted(glob.glob(os.path.join(p, '*.*'))))  # dir
+            elif os.path.isfile(p):
+                files.append(p)  # files
+            else:
+                raise FileNotFoundError(f'{p} does not exist')
+
+        images = [x for x in files if x.split('.')[-1].lower() in IMG_FORMATS]
+        videos = [x for x in files if x.split('.')[-1].lower() in VID_FORMATS]
+        ni, nv = len(images), len(videos)
+
+        self.img_size = img_size
+        self.stride = stride
+        self.files = images + videos
+        self.nf = ni + nv  # number of files
+        self.video_flag = [False] * ni + [True] * nv
+        self.mode = 'image'
+        self.auto = auto
+        self.transforms = transforms  # optional
+        self.vid_stride = vid_stride  # video frame-rate stride
+        if any(videos):
+            self._new_video(videos[0])  # new video
+        else:
+            self.cap = None
+        assert self.nf > 0, f'No images or videos found in {p}. ' \
+                            f'Supported formats are:\nimages: {IMG_FORMATS}\nvideos: {VID_FORMATS}'
+
+    def __iter__(self):
+        self.count = 0
+        return self
+
+    def __next__(self):
+        if self.count == self.nf:
+            raise StopIteration
+        path = self.files[self.count]
+
+        if self.video_flag[self.count]:
+            # Read video
+            self.mode = 'video'
+            for _ in range(self.vid_stride):
+                self.cap.grab()
+            ret_val, im0 = self.cap.retrieve()
+            while not ret_val:
+                self.count += 1
+                self.cap.release()
+                if self.count == self.nf:  # last video
+                    raise StopIteration
+                path = self.files[self.count]
+                self._new_video(path)
+                ret_val, im0 = self.cap.read()
+
+            self.frame += 1
+            # im0 = self._cv2_rotate(im0)  # for use if cv2 autorotation is False
+            s = f'video {self.count + 1}/{self.nf} ({self.frame}/{self.frames}) {path}: '
+
+        else:
+            # Read image
+            self.count += 1
+            im0 = cv2.imread(path)  # BGR
+            assert im0 is not None, f'Image Not Found {path}'
+            s = f'image {self.count}/{self.nf} {path}: '
+
+        if self.transforms:
+            im = self.transforms(im0)  # transforms
+        else:
+            im = letterbox(im0, self.img_size, stride=self.stride, auto=self.auto)[0]  # padded resize
+            im = im.transpose((2, 0, 1))[::-1]  # HWC to CHW, BGR to RGB
+            im = np.ascontiguousarray(im)  # contiguous
+
+        return path, im, im0, self.cap, s
+
+    def _new_video(self, path):
+        # Create a new video capture object
+        self.frame = 0
+        self.cap = cv2.VideoCapture(path)
+        self.frames = int(self.cap.get(cv2.CAP_PROP_FRAME_COUNT) / self.vid_stride)
+        self.orientation = int(self.cap.get(cv2.CAP_PROP_ORIENTATION_META))  # rotation degrees
+        # self.cap.set(cv2.CAP_PROP_ORIENTATION_AUTO, 0)  # disable https://github.com/ultralytics/yolov5/issues/8493
+
+    def _cv2_rotate(self, im):
+        # Rotate a cv2 video manually
+        if self.orientation == 0:
+            return cv2.rotate(im, cv2.ROTATE_90_CLOCKWISE)
+        elif self.orientation == 180:
+            return cv2.rotate(im, cv2.ROTATE_90_COUNTERCLOCKWISE)
+        elif self.orientation == 90:
+            return cv2.rotate(im, cv2.ROTATE_180)
+        return im
+
+    def __len__(self):
+        return self.nf  # number of files
+
+
+class LoadStreams:
+    # YOLOv5 streamloader, i.e. `python detect.py --source 'rtsp://example.com/media.mp4'  # RTSP, RTMP, HTTP streams`
+    def __init__(self, sources='file.streams', img_size=640, stride=32, auto=True, transforms=None, vid_stride=1):
+        torch.backends.cudnn.benchmark = True  # faster for fixed-size inference
+        self.mode = 'stream'
+        self.img_size = img_size
+        self.stride = stride
+        self.vid_stride = vid_stride  # video frame-rate stride
+        sources = Path(sources).read_text().rsplit() if os.path.isfile(sources) else [sources]
+        n = len(sources)
+        self.sources = [clean_str(x) for x in sources]  # clean source names for later
+        self.imgs, self.fps, self.frames, self.threads = [None] * n, [0] * n, [0] * n, [None] * n
+        for i, s in enumerate(sources):  # index, source
+            # Start thread to read frames from video stream
+            st = f'{i + 1}/{n}: {s}... '
+            if urlparse(s).hostname in ('www.youtube.com', 'youtube.com', 'youtu.be'):  # if source is YouTube video
+                # YouTube format i.e. 'https://www.youtube.com/watch?v=Zgi9g1ksQHc' or 'https://youtu.be/LNwODJXcvt4'
+                check_requirements(('pafy', 'youtube_dl==2020.12.2'))
+                import pafy
+                s = pafy.new(s).getbest(preftype='mp4').url  # YouTube URL
+            s = eval(s) if s.isnumeric() else s  # i.e. s = '0' local webcam
+            if s == 0:
+                assert not is_colab(), '--source 0 webcam unsupported on Colab. Rerun command in a local environment.'
+                assert not is_kaggle(), '--source 0 webcam unsupported on Kaggle. Rerun command in a local environment.'
+            cap = cv2.VideoCapture(s)
+            assert cap.isOpened(), f'{st}Failed to open {s}'
+            w = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
+            h = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
+            fps = cap.get(cv2.CAP_PROP_FPS)  # warning: may return 0 or nan
+            self.frames[i] = max(int(cap.get(cv2.CAP_PROP_FRAME_COUNT)), 0) or float('inf')  # infinite stream fallback
+            self.fps[i] = max((fps if math.isfinite(fps) else 0) % 100, 0) or 30  # 30 FPS fallback
+
+            _, self.imgs[i] = cap.read()  # guarantee first frame
+            self.threads[i] = Thread(target=self.update, args=([i, cap, s]), daemon=True)
+            LOGGER.info(f'{st} Success ({self.frames[i]} frames {w}x{h} at {self.fps[i]:.2f} FPS)')
+            self.threads[i].start()
+        LOGGER.info('')  # newline
+
+        # check for common shapes
+        s = np.stack([letterbox(x, img_size, stride=stride, auto=auto)[0].shape for x in self.imgs])
+        self.rect = np.unique(s, axis=0).shape[0] == 1  # rect inference if all shapes equal
+        self.auto = auto and self.rect
+        self.transforms = transforms  # optional
+        if not self.rect:
+            LOGGER.warning('WARNING ⚠️ Stream shapes differ. For optimal performance supply similarly-shaped streams.')
+
+    def update(self, i, cap, stream):
+        # Read stream `i` frames in daemon thread
+        n, f = 0, self.frames[i]  # frame number, frame array
+        while cap.isOpened() and n < f:
+            n += 1
+            cap.grab()  # .read() = .grab() followed by .retrieve()
+            if n % self.vid_stride == 0:
+                success, im = cap.retrieve()
+                if success:
+                    self.imgs[i] = im
+                else:
+                    LOGGER.warning('WARNING ⚠️ Video stream unresponsive, please check your IP camera connection.')
+                    self.imgs[i] = np.zeros_like(self.imgs[i])
+                    cap.open(stream)  # re-open stream if signal was lost
+            time.sleep(0.0)  # wait time
+
+    def __iter__(self):
+        self.count = -1
+        return self
+
+    def __next__(self):
+        self.count += 1
+        if not all(x.is_alive() for x in self.threads) or cv2.waitKey(1) == ord('q'):  # q to quit
+            cv2.destroyAllWindows()
+            raise StopIteration
+
+        im0 = self.imgs.copy()
+        if self.transforms:
+            im = np.stack([self.transforms(x) for x in im0])  # transforms
+        else:
+            im = np.stack([letterbox(x, self.img_size, stride=self.stride, auto=self.auto)[0] for x in im0])  # resize
+            im = im[..., ::-1].transpose((0, 3, 1, 2))  # BGR to RGB, BHWC to BCHW
+            im = np.ascontiguousarray(im)  # contiguous
+
+        return self.sources, im, im0, None, ''
+
+    def __len__(self):
+        return len(self.sources)  # 1E12 frames = 32 streams at 30 FPS for 30 years
+
+
+def img2label_paths(img_paths):
+    # Define label paths as a function of image paths
+    sa, sb = f'{os.sep}images{os.sep}', f'{os.sep}labels{os.sep}'  # /images/, /labels/ substrings
+    return [sb.join(x.rsplit(sa, 1)).rsplit('.', 1)[0] + '.txt' for x in img_paths]
+
+
+class LoadImagesAndLabels(Dataset):
+    # YOLOv5 train_loader/val_loader, loads images and labels for training and validation
+    cache_version = 0.6  # dataset labels *.cache version
+    rand_interp_methods = [cv2.INTER_NEAREST, cv2.INTER_LINEAR, cv2.INTER_CUBIC, cv2.INTER_AREA, cv2.INTER_LANCZOS4]
+
+    def __init__(self,
+                 path,
+                 img_size=640,
+                 batch_size=16,
+                 augment=False,
+                 hyp=None,
+                 rect=False,
+                 image_weights=False,
+                 cache_images=False,
+                 single_cls=False,
+                 stride=32,
+                 pad=0.0,
+                 min_items=0,
+                 prefix=''):
+        self.img_size = img_size
+        self.augment = augment
+        self.hyp = hyp
+        self.image_weights = image_weights
+        self.rect = False if image_weights else rect
+        self.mosaic = self.augment and not self.rect  # load 4 images at a time into a mosaic (only during training)
+        self.mosaic_border = [-img_size // 2, -img_size // 2]
+        self.stride = stride
+        self.path = path
+        self.albumentations = Albumentations(size=img_size) if augment else None
+
+        try:
+            f = []  # image files
+            for p in path if isinstance(path, list) else [path]:
+                p = Path(p)  # os-agnostic
+                if p.is_dir():  # dir
+                    f += glob.glob(str(p / '**' / '*.*'), recursive=True)
+                    # f = list(p.rglob('*.*'))  # pathlib
+                elif p.is_file():  # file
+                    with open(p) as t:
+                        t = t.read().strip().splitlines()
+                        parent = str(p.parent) + os.sep
+                        f += [x.replace('./', parent, 1) if x.startswith('./') else x for x in t]  # to global path
+                        # f += [p.parent / x.lstrip(os.sep) for x in t]  # to global path (pathlib)
+                else:
+                    raise FileNotFoundError(f'{prefix}{p} does not exist')
+            self.im_files = sorted(x.replace('/', os.sep) for x in f if x.split('.')[-1].lower() in IMG_FORMATS)
+            # self.img_files = sorted([x for x in f if x.suffix[1:].lower() in IMG_FORMATS])  # pathlib
+            assert self.im_files, f'{prefix}No images found'
+        except Exception as e:
+            raise Exception(f'{prefix}Error loading data from {path}: {e}\n{HELP_URL}') from e
+
+        # Check cache
+        self.label_files = img2label_paths(self.im_files)  # labels
+        cache_path = (p if p.is_file() else Path(self.label_files[0]).parent).with_suffix('.cache')
+        try:
+            cache, exists = np.load(cache_path, allow_pickle=True).item(), True  # load dict
+            assert cache['version'] == self.cache_version  # matches current version
+            assert cache['hash'] == get_hash(self.label_files + self.im_files)  # identical hash
+        except Exception:
+            cache, exists = self.cache_labels(cache_path, prefix), False  # run cache ops
+
+        # Display cache
+        nf, nm, ne, nc, n = cache.pop('results')  # found, missing, empty, corrupt, total
+        if exists and LOCAL_RANK in {-1, 0}:
+            d = f'Scanning {cache_path}... {nf} images, {nm + ne} backgrounds, {nc} corrupt'
+            tqdm(None, desc=prefix + d, total=n, initial=n, bar_format=TQDM_BAR_FORMAT)  # display cache results
+            if cache['msgs']:
+                LOGGER.info('\n'.join(cache['msgs']))  # display warnings
+        assert nf > 0 or not augment, f'{prefix}No labels found in {cache_path}, can not start training. {HELP_URL}'
+
+        # Read cache
+        [cache.pop(k) for k in ('hash', 'version', 'msgs')]  # remove items
+        labels, shapes, self.segments = zip(*cache.values())
+        nl = len(np.concatenate(labels, 0))  # number of labels
+        assert nl > 0 or not augment, f'{prefix}All labels empty in {cache_path}, can not start training. {HELP_URL}'
+        self.labels = list(labels)
+        self.shapes = np.array(shapes)
+        self.im_files = list(cache.keys())  # update
+        self.label_files = img2label_paths(cache.keys())  # update
+
+        # Filter images
+        if min_items:
+            include = np.array([len(x) >= min_items for x in self.labels]).nonzero()[0].astype(int)
+            LOGGER.info(f'{prefix}{n - len(include)}/{n} images filtered from dataset')
+            self.im_files = [self.im_files[i] for i in include]
+            self.label_files = [self.label_files[i] for i in include]
+            self.labels = [self.labels[i] for i in include]
+            self.segments = [self.segments[i] for i in include]
+            self.shapes = self.shapes[include]  # wh
+
+        # Create indices
+        n = len(self.shapes)  # number of images
+        bi = np.floor(np.arange(n) / batch_size).astype(int)  # batch index
+        nb = bi[-1] + 1  # number of batches
+        self.batch = bi  # batch index of image
+        self.n = n
+        self.indices = range(n)
+
+        # Update labels
+        include_class = []  # filter labels to include only these classes (optional)
+        self.segments = list(self.segments)
+        include_class_array = np.array(include_class).reshape(1, -1)
+        for i, (label, segment) in enumerate(zip(self.labels, self.segments)):
+            if include_class:
+                j = (label[:, 0:1] == include_class_array).any(1)
+                self.labels[i] = label[j]
+                if segment:
+                    self.segments[i] = [segment[idx] for idx, elem in enumerate(j) if elem]
+            if single_cls:  # single-class training, merge all classes into 0
+                self.labels[i][:, 0] = 0
+
+        # Rectangular Training
+        if self.rect:
+            # Sort by aspect ratio
+            s = self.shapes  # wh
+            ar = s[:, 1] / s[:, 0]  # aspect ratio
+            irect = ar.argsort()
+            self.im_files = [self.im_files[i] for i in irect]
+            self.label_files = [self.label_files[i] for i in irect]
+            self.labels = [self.labels[i] for i in irect]
+            self.segments = [self.segments[i] for i in irect]
+            self.shapes = s[irect]  # wh
+            ar = ar[irect]
+
+            # Set training image shapes
+            shapes = [[1, 1]] * nb
+            for i in range(nb):
+                ari = ar[bi == i]
+                mini, maxi = ari.min(), ari.max()
+                if maxi < 1:
+                    shapes[i] = [maxi, 1]
+                elif mini > 1:
+                    shapes[i] = [1, 1 / mini]
+
+            self.batch_shapes = np.ceil(np.array(shapes) * img_size / stride + pad).astype(int) * stride
+
+        # Cache images into RAM/disk for faster training
+        if cache_images == 'ram' and not self.check_cache_ram(prefix=prefix):
+            cache_images = False
+        self.ims = [None] * n
+        self.npy_files = [Path(f).with_suffix('.npy') for f in self.im_files]
+        if cache_images:
+            b, gb = 0, 1 << 30  # bytes of cached images, bytes per gigabytes
+            self.im_hw0, self.im_hw = [None] * n, [None] * n
+            fcn = self.cache_images_to_disk if cache_images == 'disk' else self.load_image
+            results = ThreadPool(NUM_THREADS).imap(fcn, range(n))
+            pbar = tqdm(enumerate(results), total=n, bar_format=TQDM_BAR_FORMAT, disable=LOCAL_RANK > 0)
+            for i, x in pbar:
+                if cache_images == 'disk':
+                    b += self.npy_files[i].stat().st_size
+                else:  # 'ram'
+                    self.ims[i], self.im_hw0[i], self.im_hw[i] = x  # im, hw_orig, hw_resized = load_image(self, i)
+                    b += self.ims[i].nbytes
+                pbar.desc = f'{prefix}Caching images ({b / gb:.1f}GB {cache_images})'
+            pbar.close()
+
+    def check_cache_ram(self, safety_margin=0.1, prefix=''):
+        # Check image caching requirements vs available memory
+        b, gb = 0, 1 << 30  # bytes of cached images, bytes per gigabytes
+        n = min(self.n, 30)  # extrapolate from 30 random images
+        for _ in range(n):
+            im = cv2.imread(random.choice(self.im_files))  # sample image
+            ratio = self.img_size / max(im.shape[0], im.shape[1])  # max(h, w)  # ratio
+            b += im.nbytes * ratio ** 2
+        mem_required = b * self.n / n  # GB required to cache dataset into RAM
+        mem = psutil.virtual_memory()
+        cache = mem_required * (1 + safety_margin) < mem.available  # to cache or not to cache, that is the question
+        if not cache:
+            LOGGER.info(f'{prefix}{mem_required / gb:.1f}GB RAM required, '
+                        f'{mem.available / gb:.1f}/{mem.total / gb:.1f}GB available, '
+                        f"{'caching images ✅' if cache else 'not caching images ⚠️'}")
+        return cache
+
+    def cache_labels(self, path=Path('./labels.cache'), prefix=''):
+        # Cache dataset labels, check images and read shapes
+        x = {}  # dict
+        nm, nf, ne, nc, msgs = 0, 0, 0, 0, []  # number missing, found, empty, corrupt, messages
+        desc = f'{prefix}Scanning {path.parent / path.stem}...'
+        with Pool(NUM_THREADS) as pool:
+            pbar = tqdm(pool.imap(verify_image_label, zip(self.im_files, self.label_files, repeat(prefix))),
+                        desc=desc,
+                        total=len(self.im_files),
+                        bar_format=TQDM_BAR_FORMAT)
+            for im_file, lb, shape, segments, nm_f, nf_f, ne_f, nc_f, msg in pbar:
+                nm += nm_f
+                nf += nf_f
+                ne += ne_f
+                nc += nc_f
+                if im_file:
+                    x[im_file] = [lb, shape, segments]
+                if msg:
+                    msgs.append(msg)
+                pbar.desc = f'{desc} {nf} images, {nm + ne} backgrounds, {nc} corrupt'
+
+        pbar.close()
+        if msgs:
+            LOGGER.info('\n'.join(msgs))
+        if nf == 0:
+            LOGGER.warning(f'{prefix}WARNING ⚠️ No labels found in {path}. {HELP_URL}')
+        x['hash'] = get_hash(self.label_files + self.im_files)
+        x['results'] = nf, nm, ne, nc, len(self.im_files)
+        x['msgs'] = msgs  # warnings
+        x['version'] = self.cache_version  # cache version
+        try:
+            np.save(path, x)  # save cache for next time
+            path.with_suffix('.cache.npy').rename(path)  # remove .npy suffix
+            LOGGER.info(f'{prefix}New cache created: {path}')
+        except Exception as e:
+            LOGGER.warning(f'{prefix}WARNING ⚠️ Cache directory {path.parent} is not writeable: {e}')  # not writeable
+        return x
+
+    def __len__(self):
+        return len(self.im_files)
+
+    # def __iter__(self):
+    #     self.count = -1
+    #     print('ran dataset iter')
+    #     #self.shuffled_vector = np.random.permutation(self.nF) if self.augment else np.arange(self.nF)
+    #     return self
+
+    def __getitem__(self, index):
+        index = self.indices[index]  # linear, shuffled, or image_weights
+
+        hyp = self.hyp
+        mosaic = self.mosaic and random.random() < hyp['mosaic']
+        if mosaic:
+            # Load mosaic
+            img, labels = self.load_mosaic(index)
+            shapes = None
+
+            # MixUp augmentation
+            if random.random() < hyp['mixup']:
+                img, labels = mixup(img, labels, *self.load_mosaic(random.randint(0, self.n - 1)))
+
+        else:
+            # Load image
+            img, (h0, w0), (h, w) = self.load_image(index)
+
+            # Letterbox
+            shape = self.batch_shapes[self.batch[index]] if self.rect else self.img_size  # final letterboxed shape
+            img, ratio, pad = letterbox(img, shape, auto=False, scaleup=self.augment)
+            shapes = (h0, w0), ((h / h0, w / w0), pad)  # for COCO mAP rescaling
+
+            labels = self.labels[index].copy()
+            if labels.size:  # normalized xywh to pixel xyxy format
+                labels[:, 1:] = xywhn2xyxy(labels[:, 1:], ratio[0] * w, ratio[1] * h, padw=pad[0], padh=pad[1])
+
+            if self.augment:
+                img, labels = random_perspective(img,
+                                                 labels,
+                                                 degrees=hyp['degrees'],
+                                                 translate=hyp['translate'],
+                                                 scale=hyp['scale'],
+                                                 shear=hyp['shear'],
+                                                 perspective=hyp['perspective'])
+
+        nl = len(labels)  # number of labels
+        if nl:
+            labels[:, 1:5] = xyxy2xywhn(labels[:, 1:5], w=img.shape[1], h=img.shape[0], clip=True, eps=1E-3)
+
+        if self.augment:
+            # Albumentations
+            img, labels = self.albumentations(img, labels)
+            nl = len(labels)  # update after albumentations
+
+            # HSV color-space
+            augment_hsv(img, hgain=hyp['hsv_h'], sgain=hyp['hsv_s'], vgain=hyp['hsv_v'])
+
+            # Flip up-down
+            if random.random() < hyp['flipud']:
+                img = np.flipud(img)
+                if nl:
+                    labels[:, 2] = 1 - labels[:, 2]
+
+            # Flip left-right
+            if random.random() < hyp['fliplr']:
+                img = np.fliplr(img)
+                if nl:
+                    labels[:, 1] = 1 - labels[:, 1]
+
+            # Cutouts
+            # labels = cutout(img, labels, p=0.5)
+            # nl = len(labels)  # update after cutout
+
+        labels_out = torch.zeros((nl, 6))
+        if nl:
+            labels_out[:, 1:] = torch.from_numpy(labels)
+
+        # Convert
+        img = img.transpose((2, 0, 1))[::-1]  # HWC to CHW, BGR to RGB
+        img = np.ascontiguousarray(img)
+
+        return torch.from_numpy(img), labels_out, self.im_files[index], shapes
+
+    def load_image(self, i):
+        # Loads 1 image from dataset index 'i', returns (im, original hw, resized hw)
+        im, f, fn = self.ims[i], self.im_files[i], self.npy_files[i],
+        if im is None:  # not cached in RAM
+            if fn.exists():  # load npy
+                im = np.load(fn)
+            else:  # read image
+                im = cv2.imread(f)  # BGR
+                assert im is not None, f'Image Not Found {f}'
+            h0, w0 = im.shape[:2]  # orig hw
+            r = self.img_size / max(h0, w0)  # ratio
+            if r != 1:  # if sizes are not equal
+                interp = cv2.INTER_LINEAR if (self.augment or r > 1) else cv2.INTER_AREA
+                im = cv2.resize(im, (math.ceil(w0 * r), math.ceil(h0 * r)), interpolation=interp)
+            return im, (h0, w0), im.shape[:2]  # im, hw_original, hw_resized
+        return self.ims[i], self.im_hw0[i], self.im_hw[i]  # im, hw_original, hw_resized
+
+    def cache_images_to_disk(self, i):
+        # Saves an image as an *.npy file for faster loading
+        f = self.npy_files[i]
+        if not f.exists():
+            np.save(f.as_posix(), cv2.imread(self.im_files[i]))
+
+    def load_mosaic(self, index):
+        # YOLOv5 4-mosaic loader. Loads 1 image + 3 random images into a 4-image mosaic
+        labels4, segments4 = [], []
+        s = self.img_size
+        yc, xc = (int(random.uniform(-x, 2 * s + x)) for x in self.mosaic_border)  # mosaic center x, y
+        indices = [index] + random.choices(self.indices, k=3)  # 3 additional image indices
+        random.shuffle(indices)
+        for i, index in enumerate(indices):
+            # Load image
+            img, _, (h, w) = self.load_image(index)
+
+            # place img in img4
+            if i == 0:  # top left
+                img4 = np.full((s * 2, s * 2, img.shape[2]), 114, dtype=np.uint8)  # base image with 4 tiles
+                x1a, y1a, x2a, y2a = max(xc - w, 0), max(yc - h, 0), xc, yc  # xmin, ymin, xmax, ymax (large image)
+                x1b, y1b, x2b, y2b = w - (x2a - x1a), h - (y2a - y1a), w, h  # xmin, ymin, xmax, ymax (small image)
+            elif i == 1:  # top right
+                x1a, y1a, x2a, y2a = xc, max(yc - h, 0), min(xc + w, s * 2), yc
+                x1b, y1b, x2b, y2b = 0, h - (y2a - y1a), min(w, x2a - x1a), h
+            elif i == 2:  # bottom left
+                x1a, y1a, x2a, y2a = max(xc - w, 0), yc, xc, min(s * 2, yc + h)
+                x1b, y1b, x2b, y2b = w - (x2a - x1a), 0, w, min(y2a - y1a, h)
+            elif i == 3:  # bottom right
+                x1a, y1a, x2a, y2a = xc, yc, min(xc + w, s * 2), min(s * 2, yc + h)
+                x1b, y1b, x2b, y2b = 0, 0, min(w, x2a - x1a), min(y2a - y1a, h)
+
+            img4[y1a:y2a, x1a:x2a] = img[y1b:y2b, x1b:x2b]  # img4[ymin:ymax, xmin:xmax]
+            padw = x1a - x1b
+            padh = y1a - y1b
+
+            # Labels
+            labels, segments = self.labels[index].copy(), self.segments[index].copy()
+            if labels.size:
+                labels[:, 1:] = xywhn2xyxy(labels[:, 1:], w, h, padw, padh)  # normalized xywh to pixel xyxy format
+                segments = [xyn2xy(x, w, h, padw, padh) for x in segments]
+            labels4.append(labels)
+            segments4.extend(segments)
+
+        # Concat/clip labels
+        labels4 = np.concatenate(labels4, 0)
+        for x in (labels4[:, 1:], *segments4):
+            np.clip(x, 0, 2 * s, out=x)  # clip when using random_perspective()
+        # img4, labels4 = replicate(img4, labels4)  # replicate
+
+        # Augment
+        img4, labels4, segments4 = copy_paste(img4, labels4, segments4, p=self.hyp['copy_paste'])
+        img4, labels4 = random_perspective(img4,
+                                           labels4,
+                                           segments4,
+                                           degrees=self.hyp['degrees'],
+                                           translate=self.hyp['translate'],
+                                           scale=self.hyp['scale'],
+                                           shear=self.hyp['shear'],
+                                           perspective=self.hyp['perspective'],
+                                           border=self.mosaic_border)  # border to remove
+
+        return img4, labels4
+
+    def load_mosaic9(self, index):
+        # YOLOv5 9-mosaic loader. Loads 1 image + 8 random images into a 9-image mosaic
+        labels9, segments9 = [], []
+        s = self.img_size
+        indices = [index] + random.choices(self.indices, k=8)  # 8 additional image indices
+        random.shuffle(indices)
+        hp, wp = -1, -1  # height, width previous
+        for i, index in enumerate(indices):
+            # Load image
+            img, _, (h, w) = self.load_image(index)
+
+            # place img in img9
+            if i == 0:  # center
+                img9 = np.full((s * 3, s * 3, img.shape[2]), 114, dtype=np.uint8)  # base image with 4 tiles
+                h0, w0 = h, w
+                c = s, s, s + w, s + h  # xmin, ymin, xmax, ymax (base) coordinates
+            elif i == 1:  # top
+                c = s, s - h, s + w, s
+            elif i == 2:  # top right
+                c = s + wp, s - h, s + wp + w, s
+            elif i == 3:  # right
+                c = s + w0, s, s + w0 + w, s + h
+            elif i == 4:  # bottom right
+                c = s + w0, s + hp, s + w0 + w, s + hp + h
+            elif i == 5:  # bottom
+                c = s + w0 - w, s + h0, s + w0, s + h0 + h
+            elif i == 6:  # bottom left
+                c = s + w0 - wp - w, s + h0, s + w0 - wp, s + h0 + h
+            elif i == 7:  # left
+                c = s - w, s + h0 - h, s, s + h0
+            elif i == 8:  # top left
+                c = s - w, s + h0 - hp - h, s, s + h0 - hp
+
+            padx, pady = c[:2]
+            x1, y1, x2, y2 = (max(x, 0) for x in c)  # allocate coords
+
+            # Labels
+            labels, segments = self.labels[index].copy(), self.segments[index].copy()
+            if labels.size:
+                labels[:, 1:] = xywhn2xyxy(labels[:, 1:], w, h, padx, pady)  # normalized xywh to pixel xyxy format
+                segments = [xyn2xy(x, w, h, padx, pady) for x in segments]
+            labels9.append(labels)
+            segments9.extend(segments)
+
+            # Image
+            img9[y1:y2, x1:x2] = img[y1 - pady:, x1 - padx:]  # img9[ymin:ymax, xmin:xmax]
+            hp, wp = h, w  # height, width previous
+
+        # Offset
+        yc, xc = (int(random.uniform(0, s)) for _ in self.mosaic_border)  # mosaic center x, y
+        img9 = img9[yc:yc + 2 * s, xc:xc + 2 * s]
+
+        # Concat/clip labels
+        labels9 = np.concatenate(labels9, 0)
+        labels9[:, [1, 3]] -= xc
+        labels9[:, [2, 4]] -= yc
+        c = np.array([xc, yc])  # centers
+        segments9 = [x - c for x in segments9]
+
+        for x in (labels9[:, 1:], *segments9):
+            np.clip(x, 0, 2 * s, out=x)  # clip when using random_perspective()
+        # img9, labels9 = replicate(img9, labels9)  # replicate
+
+        # Augment
+        img9, labels9, segments9 = copy_paste(img9, labels9, segments9, p=self.hyp['copy_paste'])
+        img9, labels9 = random_perspective(img9,
+                                           labels9,
+                                           segments9,
+                                           degrees=self.hyp['degrees'],
+                                           translate=self.hyp['translate'],
+                                           scale=self.hyp['scale'],
+                                           shear=self.hyp['shear'],
+                                           perspective=self.hyp['perspective'],
+                                           border=self.mosaic_border)  # border to remove
+
+        return img9, labels9
+
+    @staticmethod
+    def collate_fn(batch):
+        im, label, path, shapes = zip(*batch)  # transposed
+        for i, lb in enumerate(label):
+            lb[:, 0] = i  # add target image index for build_targets()
+        return torch.stack(im, 0), torch.cat(label, 0), path, shapes
+
+    @staticmethod
+    def collate_fn4(batch):
+        im, label, path, shapes = zip(*batch)  # transposed
+        n = len(shapes) // 4
+        im4, label4, path4, shapes4 = [], [], path[:n], shapes[:n]
+
+        ho = torch.tensor([[0.0, 0, 0, 1, 0, 0]])
+        wo = torch.tensor([[0.0, 0, 1, 0, 0, 0]])
+        s = torch.tensor([[1, 1, 0.5, 0.5, 0.5, 0.5]])  # scale
+        for i in range(n):  # zidane torch.zeros(16,3,720,1280)  # BCHW
+            i *= 4
+            if random.random() < 0.5:
+                im1 = F.interpolate(im[i].unsqueeze(0).float(), scale_factor=2.0, mode='bilinear',
+                                    align_corners=False)[0].type(im[i].type())
+                lb = label[i]
+            else:
+                im1 = torch.cat((torch.cat((im[i], im[i + 1]), 1), torch.cat((im[i + 2], im[i + 3]), 1)), 2)
+                lb = torch.cat((label[i], label[i + 1] + ho, label[i + 2] + wo, label[i + 3] + ho + wo), 0) * s
+            im4.append(im1)
+            label4.append(lb)
+
+        for i, lb in enumerate(label4):
+            lb[:, 0] = i  # add target image index for build_targets()
+
+        return torch.stack(im4, 0), torch.cat(label4, 0), path4, shapes4
+
+
+# Ancillary functions --------------------------------------------------------------------------------------------------
+def flatten_recursive(path=DATASETS_DIR / 'coco128'):
+    # Flatten a recursive directory by bringing all files to top level
+    new_path = Path(f'{str(path)}_flat')
+    if os.path.exists(new_path):
+        shutil.rmtree(new_path)  # delete output folder
+    os.makedirs(new_path)  # make new output folder
+    for file in tqdm(glob.glob(f'{str(Path(path))}/**/*.*', recursive=True)):
+        shutil.copyfile(file, new_path / Path(file).name)
+
+
+def extract_boxes(path=DATASETS_DIR / 'coco128'):  # from utils.dataloaders import *; extract_boxes()
+    # Convert detection dataset into classification dataset, with one directory per class
+    path = Path(path)  # images dir
+    shutil.rmtree(path / 'classification') if (path / 'classification').is_dir() else None  # remove existing
+    files = list(path.rglob('*.*'))
+    n = len(files)  # number of files
+    for im_file in tqdm(files, total=n):
+        if im_file.suffix[1:] in IMG_FORMATS:
+            # image
+            im = cv2.imread(str(im_file))[..., ::-1]  # BGR to RGB
+            h, w = im.shape[:2]
+
+            # labels
+            lb_file = Path(img2label_paths([str(im_file)])[0])
+            if Path(lb_file).exists():
+                with open(lb_file) as f:
+                    lb = np.array([x.split() for x in f.read().strip().splitlines()], dtype=np.float32)  # labels
+
+                for j, x in enumerate(lb):
+                    c = int(x[0])  # class
+                    f = (path / 'classifier') / f'{c}' / f'{path.stem}_{im_file.stem}_{j}.jpg'  # new filename
+                    if not f.parent.is_dir():
+                        f.parent.mkdir(parents=True)
+
+                    b = x[1:] * [w, h, w, h]  # box
+                    # b[2:] = b[2:].max()  # rectangle to square
+                    b[2:] = b[2:] * 1.2 + 3  # pad
+                    b = xywh2xyxy(b.reshape(-1, 4)).ravel().astype(int)
+
+                    b[[0, 2]] = np.clip(b[[0, 2]], 0, w)  # clip boxes outside of image
+                    b[[1, 3]] = np.clip(b[[1, 3]], 0, h)
+                    assert cv2.imwrite(str(f), im[b[1]:b[3], b[0]:b[2]]), f'box failure in {f}'
+
+
+def autosplit(path=DATASETS_DIR / 'coco128/images', weights=(0.9, 0.1, 0.0), annotated_only=False):
+    """ Autosplit a dataset into train/val/test splits and save path/autosplit_*.txt files
+    Usage: from utils.dataloaders import *; autosplit()
+    Arguments
+        path:            Path to images directory
+        weights:         Train, val, test weights (list, tuple)
+        annotated_only:  Only use images with an annotated txt file
+    """
+    path = Path(path)  # images dir
+    files = sorted(x for x in path.rglob('*.*') if x.suffix[1:].lower() in IMG_FORMATS)  # image files only
+    n = len(files)  # number of files
+    random.seed(0)  # for reproducibility
+    indices = random.choices([0, 1, 2], weights=weights, k=n)  # assign each image to a split
+
+    txt = ['autosplit_train.txt', 'autosplit_val.txt', 'autosplit_test.txt']  # 3 txt files
+    for x in txt:
+        if (path.parent / x).exists():
+            (path.parent / x).unlink()  # remove existing
+
+    print(f'Autosplitting images from {path}' + ', using *.txt labeled images only' * annotated_only)
+    for i, img in tqdm(zip(indices, files), total=n):
+        if not annotated_only or Path(img2label_paths([str(img)])[0]).exists():  # check label
+            with open(path.parent / txt[i], 'a') as f:
+                f.write(f'./{img.relative_to(path.parent).as_posix()}' + '\n')  # add image to txt file
+
+
+def verify_image_label(args):
+    # Verify one image-label pair
+    im_file, lb_file, prefix = args
+    nm, nf, ne, nc, msg, segments = 0, 0, 0, 0, '', []  # number (missing, found, empty, corrupt), message, segments
+    try:
+        # verify images
+        im = Image.open(im_file)
+        im.verify()  # PIL verify
+        shape = exif_size(im)  # image size
+        assert (shape[0] > 9) & (shape[1] > 9), f'image size {shape} <10 pixels'
+        assert im.format.lower() in IMG_FORMATS, f'invalid image format {im.format}'
+        if im.format.lower() in ('jpg', 'jpeg'):
+            with open(im_file, 'rb') as f:
+                f.seek(-2, 2)
+                if f.read() != b'\xff\xd9':  # corrupt JPEG
+                    ImageOps.exif_transpose(Image.open(im_file)).save(im_file, 'JPEG', subsampling=0, quality=100)
+                    msg = f'{prefix}WARNING ⚠️ {im_file}: corrupt JPEG restored and saved'
+
+        # verify labels
+        if os.path.isfile(lb_file):
+            nf = 1  # label found
+            with open(lb_file) as f:
+                lb = [x.split() for x in f.read().strip().splitlines() if len(x)]
+                if any(len(x) > 6 for x in lb):  # is segment
+                    classes = np.array([x[0] for x in lb], dtype=np.float32)
+                    segments = [np.array(x[1:], dtype=np.float32).reshape(-1, 2) for x in lb]  # (cls, xy1...)
+                    lb = np.concatenate((classes.reshape(-1, 1), segments2boxes(segments)), 1)  # (cls, xywh)
+                lb = np.array(lb, dtype=np.float32)
+            nl = len(lb)
+            if nl:
+                assert lb.shape[1] == 5, f'labels require 5 columns, {lb.shape[1]} columns detected'
+                assert (lb >= 0).all(), f'negative label values {lb[lb < 0]}'
+                assert (lb[:, 1:] <= 1).all(), f'non-normalized or out of bounds coordinates {lb[:, 1:][lb[:, 1:] > 1]}'
+                _, i = np.unique(lb, axis=0, return_index=True)
+                if len(i) < nl:  # duplicate row check
+                    lb = lb[i]  # remove duplicates
+                    if segments:
+                        segments = [segments[x] for x in i]
+                    msg = f'{prefix}WARNING ⚠️ {im_file}: {nl - len(i)} duplicate labels removed'
+            else:
+                ne = 1  # label empty
+                lb = np.zeros((0, 5), dtype=np.float32)
+        else:
+            nm = 1  # label missing
+            lb = np.zeros((0, 5), dtype=np.float32)
+        return im_file, lb, shape, segments, nm, nf, ne, nc, msg
+    except Exception as e:
+        nc = 1
+        msg = f'{prefix}WARNING ⚠️ {im_file}: ignoring corrupt image/label: {e}'
+        return [None, None, None, None, nm, nf, ne, nc, msg]
+
+
+class HUBDatasetStats():
+    """ Class for generating HUB dataset JSON and `-hub` dataset directory
+
+    Arguments
+        path:           Path to data.yaml or data.zip (with data.yaml inside data.zip)
+        autodownload:   Attempt to download dataset if not found locally
+
+    Usage
+        from utils.dataloaders import HUBDatasetStats
+        stats = HUBDatasetStats('coco128.yaml', autodownload=True)  # usage 1
+        stats = HUBDatasetStats('path/to/coco128.zip')  # usage 2
+        stats.get_json(save=False)
+        stats.process_images()
+    """
+
+    def __init__(self, path='coco128.yaml', autodownload=False):
+        # Initialize class
+        zipped, data_dir, yaml_path = self._unzip(Path(path))
+        try:
+            with open(check_yaml(yaml_path), errors='ignore') as f:
+                data = yaml.safe_load(f)  # data dict
+                if zipped:
+                    data['path'] = data_dir
+        except Exception as e:
+            raise Exception('error/HUB/dataset_stats/yaml_load') from e
+
+        check_dataset(data, autodownload)  # download dataset if missing
+        self.hub_dir = Path(data['path'] + '-hub')
+        self.im_dir = self.hub_dir / 'images'
+        self.im_dir.mkdir(parents=True, exist_ok=True)  # makes /images
+        self.stats = {'nc': data['nc'], 'names': list(data['names'].values())}  # statistics dictionary
+        self.data = data
+
+    @staticmethod
+    def _find_yaml(dir):
+        # Return data.yaml file
+        files = list(dir.glob('*.yaml')) or list(dir.rglob('*.yaml'))  # try root level first and then recursive
+        assert files, f'No *.yaml file found in {dir}'
+        if len(files) > 1:
+            files = [f for f in files if f.stem == dir.stem]  # prefer *.yaml files that match dir name
+            assert files, f'Multiple *.yaml files found in {dir}, only 1 *.yaml file allowed'
+        assert len(files) == 1, f'Multiple *.yaml files found: {files}, only 1 *.yaml file allowed in {dir}'
+        return files[0]
+
+    def _unzip(self, path):
+        # Unzip data.zip
+        if not str(path).endswith('.zip'):  # path is data.yaml
+            return False, None, path
+        assert Path(path).is_file(), f'Error unzipping {path}, file not found'
+        unzip_file(path, path=path.parent)
+        dir = path.with_suffix('')  # dataset directory == zip name
+        assert dir.is_dir(), f'Error unzipping {path}, {dir} not found. path/to/abc.zip MUST unzip to path/to/abc/'
+        return True, str(dir), self._find_yaml(dir)  # zipped, data_dir, yaml_path
+
+    def _hub_ops(self, f, max_dim=1920):
+        # HUB ops for 1 image 'f': resize and save at reduced quality in /dataset-hub for web/app viewing
+        f_new = self.im_dir / Path(f).name  # dataset-hub image filename
+        try:  # use PIL
+            im = Image.open(f)
+            r = max_dim / max(im.height, im.width)  # ratio
+            if r < 1.0:  # image too large
+                im = im.resize((int(im.width * r), int(im.height * r)))
+            im.save(f_new, 'JPEG', quality=50, optimize=True)  # save
+        except Exception as e:  # use OpenCV
+            LOGGER.info(f'WARNING ⚠️ HUB ops PIL failure {f}: {e}')
+            im = cv2.imread(f)
+            im_height, im_width = im.shape[:2]
+            r = max_dim / max(im_height, im_width)  # ratio
+            if r < 1.0:  # image too large
+                im = cv2.resize(im, (int(im_width * r), int(im_height * r)), interpolation=cv2.INTER_AREA)
+            cv2.imwrite(str(f_new), im)
+
+    def get_json(self, save=False, verbose=False):
+        # Return dataset JSON for Ultralytics HUB
+        def _round(labels):
+            # Update labels to integer class and 6 decimal place floats
+            return [[int(c), *(round(x, 4) for x in points)] for c, *points in labels]
+
+        for split in 'train', 'val', 'test':
+            if self.data.get(split) is None:
+                self.stats[split] = None  # i.e. no test set
+                continue
+            dataset = LoadImagesAndLabels(self.data[split])  # load dataset
+            x = np.array([
+                np.bincount(label[:, 0].astype(int), minlength=self.data['nc'])
+                for label in tqdm(dataset.labels, total=dataset.n, desc='Statistics')])  # shape(128x80)
+            self.stats[split] = {
+                'instance_stats': {
+                    'total': int(x.sum()),
+                    'per_class': x.sum(0).tolist()},
+                'image_stats': {
+                    'total': dataset.n,
+                    'unlabelled': int(np.all(x == 0, 1).sum()),
+                    'per_class': (x > 0).sum(0).tolist()},
+                'labels': [{
+                    str(Path(k).name): _round(v.tolist())} for k, v in zip(dataset.im_files, dataset.labels)]}
+
+        # Save, print and return
+        if save:
+            stats_path = self.hub_dir / 'stats.json'
+            print(f'Saving {stats_path.resolve()}...')
+            with open(stats_path, 'w') as f:
+                json.dump(self.stats, f)  # save stats.json
+        if verbose:
+            print(json.dumps(self.stats, indent=2, sort_keys=False))
+        return self.stats
+
+    def process_images(self):
+        # Compress images for Ultralytics HUB
+        for split in 'train', 'val', 'test':
+            if self.data.get(split) is None:
+                continue
+            dataset = LoadImagesAndLabels(self.data[split])  # load dataset
+            desc = f'{split} images'
+            for _ in tqdm(ThreadPool(NUM_THREADS).imap(self._hub_ops, dataset.im_files), total=dataset.n, desc=desc):
+                pass
+        print(f'Done. All images saved to {self.im_dir}')
+        return self.im_dir
+
+
+# Classification dataloaders -------------------------------------------------------------------------------------------
+class ClassificationDataset(torchvision.datasets.ImageFolder):
+    """
+    YOLOv5 Classification Dataset.
+    Arguments
+        root:  Dataset path
+        transform:  torchvision transforms, used by default
+        album_transform: Albumentations transforms, used if installed
+    """
+
+    def __init__(self, root, augment, imgsz, cache=False):
+        super().__init__(root=root)
+        self.torch_transforms = classify_transforms(imgsz)
+        self.album_transforms = classify_albumentations(augment, imgsz) if augment else None
+        self.cache_ram = cache is True or cache == 'ram'
+        self.cache_disk = cache == 'disk'
+        self.samples = [list(x) + [Path(x[0]).with_suffix('.npy'), None] for x in self.samples]  # file, index, npy, im
+
+    def __getitem__(self, i):
+        f, j, fn, im = self.samples[i]  # filename, index, filename.with_suffix('.npy'), image
+        if self.cache_ram and im is None:
+            im = self.samples[i][3] = cv2.imread(f)
+        elif self.cache_disk:
+            if not fn.exists():  # load npy
+                np.save(fn.as_posix(), cv2.imread(f))
+            im = np.load(fn)
+        else:  # read image
+            im = cv2.imread(f)  # BGR
+        if self.album_transforms:
+            sample = self.album_transforms(image=cv2.cvtColor(im, cv2.COLOR_BGR2RGB))['image']
+        else:
+            sample = self.torch_transforms(im)
+        return sample, j
+
+
+def create_classification_dataloader(path,
+                                     imgsz=224,
+                                     batch_size=16,
+                                     augment=True,
+                                     cache=False,
+                                     rank=-1,
+                                     workers=8,
+                                     shuffle=True):
+    # Returns Dataloader object to be used with YOLOv5 Classifier
+    with torch_distributed_zero_first(rank):  # init dataset *.cache only once if DDP
+        dataset = ClassificationDataset(root=path, imgsz=imgsz, augment=augment, cache=cache)
+    batch_size = min(batch_size, len(dataset))
+    nd = torch.cuda.device_count()
+    nw = min([os.cpu_count() // max(nd, 1), batch_size if batch_size > 1 else 0, workers])
+    sampler = None if rank == -1 else distributed.DistributedSampler(dataset, shuffle=shuffle)
+    generator = torch.Generator()
+    generator.manual_seed(6148914691236517205 + RANK)
+    return InfiniteDataLoader(dataset,
+                              batch_size=batch_size,
+                              shuffle=shuffle and sampler is None,
+                              num_workers=nw,
+                              sampler=sampler,
+                              pin_memory=PIN_MEMORY,
+                              worker_init_fn=seed_worker,
+                              generator=generator)  # or DataLoader(persistent_workers=True)

utils/docker/Dockerfile

@@ -0,0 +1,73 @@
+# YOLOv5 🚀 by Ultralytics, AGPL-3.0 license
+# Builds ultralytics/yolov5:latest image on DockerHub https://hub.docker.com/r/ultralytics/yolov5
+# Image is CUDA-optimized for YOLOv5 single/multi-GPU training and inference
+
+# Start FROM PyTorch image https://hub.docker.com/r/pytorch/pytorch
+FROM pytorch/pytorch:2.0.0-cuda11.7-cudnn8-runtime
+
+# Downloads to user config dir
+ADD https://ultralytics.com/assets/Arial.ttf https://ultralytics.com/assets/Arial.Unicode.ttf /root/.config/Ultralytics/
+
+# Install linux packages
+ENV DEBIAN_FRONTEND noninteractive
+RUN apt update
+RUN TZ=Etc/UTC apt install -y tzdata
+RUN apt install --no-install-recommends -y gcc git zip curl htop libgl1-mesa-glx libglib2.0-0 libpython3-dev gnupg
+# RUN alias python=python3
+
+# Security updates
+# https://security.snyk.io/vuln/SNYK-UBUNTU1804-OPENSSL-3314796
+RUN apt upgrade --no-install-recommends -y openssl
+
+# Create working directory
+RUN rm -rf /usr/src/app && mkdir -p /usr/src/app
+WORKDIR /usr/src/app
+
+# Copy contents
+COPY . /usr/src/app
+
+# Install pip packages
+COPY requirements.txt .
+RUN python3 -m pip install --upgrade pip wheel
+RUN pip install --no-cache -r requirements.txt albumentations comet gsutil notebook \
+    coremltools onnx onnx-simplifier onnxruntime 'openvino-dev>=2023.0'
+    # tensorflow tensorflowjs \
+
+# Set environment variables
+ENV OMP_NUM_THREADS=1
+
+# Cleanup
+ENV DEBIAN_FRONTEND teletype
+
+
+# Usage Examples -------------------------------------------------------------------------------------------------------
+
+# Build and Push
+# t=ultralytics/yolov5:latest && sudo docker build -f utils/docker/Dockerfile -t $t . && sudo docker push $t
+
+# Pull and Run
+# t=ultralytics/yolov5:latest && sudo docker pull $t && sudo docker run -it --ipc=host --gpus all $t
+
+# Pull and Run with local directory access
+# t=ultralytics/yolov5:latest && sudo docker pull $t && sudo docker run -it --ipc=host --gpus all -v "$(pwd)"/datasets:/usr/src/datasets $t
+
+# Kill all
+# sudo docker kill $(sudo docker ps -q)
+
+# Kill all image-based
+# sudo docker kill $(sudo docker ps -qa --filter ancestor=ultralytics/yolov5:latest)
+
+# DockerHub tag update
+# t=ultralytics/yolov5:latest tnew=ultralytics/yolov5:v6.2 && sudo docker pull $t && sudo docker tag $t $tnew && sudo docker push $tnew
+
+# Clean up
+# sudo docker system prune -a --volumes
+
+# Update Ubuntu drivers
+# https://www.maketecheasier.com/install-nvidia-drivers-ubuntu/
+
+# DDP test
+# python -m torch.distributed.run --nproc_per_node 2 --master_port 1 train.py --epochs 3
+
+# GCP VM from Image
+# docker.io/ultralytics/yolov5:latest