Updated ONNX files to allow for dynamic / smaller batch sizes, added tiling and example images on LFS

.gitattributes

@@ -1,2 +1,12 @@
 *.onnx filter=lfs diff=lfs merge=lfs -text
 *.weights filter=lfs diff=lfs merge=lfs -text
+examples/0d4e4df2-7b69-91b1-1985-c8421f2f3253.jpg filter=lfs diff=lfs merge=lfs -text
+examples/18cef191-74ed-2b5e-55a5-f58bd3d483ff.jpg filter=lfs diff=lfs merge=lfs -text
+examples/1be4d40a-6fd0-42ce-da6c-294e45781f41.jpg filter=lfs diff=lfs merge=lfs -text
+examples/1d3c85e9-ee24-f290-e7e1-6e338f2eaebb.jpg filter=lfs diff=lfs merge=lfs -text
+examples/3e043302-af1c-75a7-4057-3a2f25c123bf.jpg filter=lfs diff=lfs merge=lfs -text
+examples/43ecc08d-502a-7a51-9d68-3e40a76439a2.jpg filter=lfs diff=lfs merge=lfs -text
+examples/479058af-e774-e6aa-a2b0-9a42dd6ff8b1.jpg filter=lfs diff=lfs merge=lfs -text
+examples/7c910b87-ae3a-f580-d431-03cd89793803.jpg filter=lfs diff=lfs merge=lfs -text
+examples/8fa04489-cd94-7d8f-7e2e-5f0fe2f7ae76.jpg filter=lfs diff=lfs merge=lfs -text
+examples/bb7b4345-b98a-c727-4c94-6090f0aa4355.jpg filter=lfs diff=lfs merge=lfs -text

.github/workflows/python-publish.yml

@@ -16,9 +16,6 @@ jobs:
 
     steps:
       - uses: actions/checkout@v2
-        with:
-          # This allows the setuptools_scm library to discover the tag version from git
-          fetch-depth: 0
 
       - uses: actions/setup-python@v2
         name: Install Python

.github/workflows/testing.yml

@@ -18,7 +18,10 @@ jobs:
       PYTHON: ${{ matrix.python-version }}
     steps:
       # Checkout and env setup
-      - uses: actions/checkout@v2
+      - name: Checkout code
+        uses: nschloe/action-cached-lfs-checkout@v1.1.3
+        with:
+          include: "*.0.onnx"
 
       - name: Set up Python ${{ matrix.python-version }}
         uses: actions/setup-python@v2
@@ -29,6 +32,7 @@ jobs:
         run: |
           python -m pip install --upgrade pip
           pip install -r requirements.txt
+          pip install -r requirements.optional.txt
 
       - name: Lint with flake8
         run: |

LICENSE

@@ -198,4 +198,4 @@ Apache License
    distributed under the License is distributed on an "AS IS" BASIS,
    WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
    See the License for the specific language governing permissions and
-   limitations under the License.
+   limitations under the License.

app.py

@@ -1,9 +1,9 @@
 # -*- coding: utf-8 -*-
+import cv2
 import gradio as gr
 import numpy as np
-import cv2
 
-from scoutbot import wic, loc
+from scoutbot import loc, wic
 
 
 def predict(filepath, wic_thresh, loc_thresh, nms_thresh):
@@ -44,9 +44,7 @@ def predict(filepath, wic_thresh, loc_thresh, nms_thresh):
                 )
                 color = (255, 0, 0)
                 img = cv2.rectangle(img, point1, point2, color, 2)
-                loc_detections.append(
-                    f'{detect.class_label}: {detect.confidence:0.05f}'
-                )
+                loc_detections.append(f'{detect.class_label}: {detect.confidence:0.05f}')
     loc_detections = '\n'.join(loc_detections)
 
     return img, wic_confidence, loc_detections

requirements.txt

@@ -1,13 +1,13 @@
-onnxruntime
+click
+cryptography
+gradio
+imgaug
 numpy
-wbia-utool
-torch
-torchvision
+onnxruntime
 opencv-python-headless
 Pillow
-imgaug
 rich
+torch
+torchvision
 tqdm
-gradio
-cryptography
-click
+wbia-utool

scoutbot/__init__.py

@@ -2,11 +2,6 @@
 '''
 2022 Wild Me
 '''
-from scoutbot import utils
-
 VERSION = '0.1.0'
 version = VERSION
 __version__ = VERSION
-
-
-log = utils.init_logging()

scoutbot/loc/__init__.py

@@ -3,26 +3,27 @@
 2022 Wild Me
 '''
 from os.path import join
-import onnxruntime as ort
 from pathlib import Path
-import torchvision
+
+import cv2
 import numpy as np
-import utool as ut
+import onnxruntime as ort
 import torch
-import cv2
+import torchvision
+import utool as ut
+
 from scoutbot.loc.transforms import (
-    Letterbox,
     Compose,
     GetBoundingBoxes,
+    Letterbox,
     NonMaxSupression,
-    TensorToBrambox,
     ReverseLetterbox,
+    TensorToBrambox,
 )
 
-
 PWD = Path(__file__).absolute().parent
 
-BATCH_SIZE = 128
+BATCH_SIZE = 16
 INPUT_SIZE = (416, 416)
 INPUT_SIZE_H, INPUT_SIZE_W = INPUT_SIZE
 NETWORK_SIZE = (INPUT_SIZE_H, INPUT_SIZE_W, 3)
@@ -52,10 +53,7 @@ def pre(inputs):
         size = img.shape[:2][::-1]
 
         img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
-        img = Letterbox.apply(
-            img,
-            dimension=INPUT_SIZE
-        )
+        img = Letterbox.apply(img, dimension=INPUT_SIZE)
         img = transform(img)
 
         data.append(img.tolist())
@@ -64,17 +62,17 @@ def pre(inputs):
     return data, sizes
 
 
-def predict(data):
-    ort_session = ort.InferenceSession(
-        ONNX_MODEL,
-        providers=['CPUExecutionProvider']
-    )
+def predict(data, fill=True):
+    ort_session = ort.InferenceSession(ONNX_MODEL, providers=['CPUExecutionProvider'])
 
     preds = []
     for chunk in ut.ichunks(data, BATCH_SIZE):
         trim = len(chunk)
-        while(len(chunk)) < BATCH_SIZE:
-            chunk.append(np.random.randn(3, INPUT_SIZE_H, INPUT_SIZE_W).astype(np.float32))
+        if fill:
+            while (len(chunk)) < BATCH_SIZE:
+                chunk.append(
+                    np.random.randn(3, INPUT_SIZE_H, INPUT_SIZE_W).astype(np.float32)
+                )
         input_ = np.array(chunk, dtype=np.float32)
 
         pred_ = ort_session.run(
@@ -89,9 +87,7 @@ def predict(data):
 def post(preds, sizes, loc_thresh=CONF_THRESH, nms_thresh=NMS_THRESH):
     postprocess = Compose(
         [
-            GetBoundingBoxes(
-                NUM_CLASSES, ANCHORS, loc_thresh
-            ),
+            GetBoundingBoxes(NUM_CLASSES, ANCHORS, loc_thresh),
             NonMaxSupression(nms_thresh),
             TensorToBrambox(NETWORK_SIZE, CLASS_LABEL_MAP),
         ]
@@ -101,9 +97,7 @@ def post(preds, sizes, loc_thresh=CONF_THRESH, nms_thresh=NMS_THRESH):
 
     outputs = []
     for pred, size in zip(preds, sizes):
-        output = ReverseLetterbox.apply(
-            [pred], INPUT_SIZE, size
-        )
+        output = ReverseLetterbox.apply([pred], INPUT_SIZE, size)
         outputs.append(output[0])
 
     return outputs

scoutbot/loc/convert.py

@@ -1,24 +1,7 @@
+# -*- coding: utf-8 -*-
 """
-
 pip install torch torchvision onnx onnxruntime-gpu tqdm wbia-utool scikit-learn numpy
 
-"""
-import torch
-import torchvision
-import onnx
-import onnxruntime as ort
-import tqdm
-import random
-import utool as ut
-import vtool as vt
-import cv2
-import numpy as np
-import lightnet as ln
-import sklearn
-import time
-from os.path import join, exists, split, splitext
-
-"""
 detection_config = {
     'algo': 'tile_aggregation',
     'config_filepath': 'variant3-32',
@@ -51,9 +34,28 @@ prediction_list = depc.get_property(
     'localizations', gid_list_, None, config=config
 )
 """
+import random
+import time
+from os.path import exists, join, split, splitext
+
+import cv2
+import lightnet as ln
+import numpy as np
+import onnx
+import onnxruntime as ort
+import sklearn
+import torch
+import torchvision
+import tqdm
+import utool as ut
+import vtool as vt
+import wbia
 
 WITH_GPU = False
-BATCH_SIZE = 128
+BATCH_SIZE = 16
+
+
+ibs = wbia.opendb(dbdir='/data/db')
 
 
 pkl_path = 'scout.pkl'
@@ -76,7 +78,9 @@ if not exists(pkl_path):
     for chunk_tids in tqdm.tqdm(ut.ichunks(tids, 1000)):
         _, _, chunk_flags = ibs.scout_tile_positive_cumulative_area(chunk_tids)
         chunk_filepaths = ibs.get_image_paths(chunk_tids)
-        for index, (tid, flag, filepath) in enumerate(zip(chunk_tids, chunk_flags, chunk_filepaths)):
+        for index, (tid, flag, filepath) in enumerate(
+            zip(chunk_tids, chunk_flags, chunk_filepaths)
+        ):
             if not exists(filepath):
                 continue
             if flag:
@@ -125,8 +129,8 @@ INDEX = 1
 
 config_path = f'/cache/lightnet/detect.lightnet.scout.5fbfff26.v{INDEX}.py'
 weights_path = f'/cache/lightnet/detect.lightnet.scout.5fbfff26.v{INDEX}.weights'
-conf_thresh = 0.4
-nms_thresh = 1.0
+conf_thresh = 0.0
+nms_thresh = 0.2
 
 assert exists(config_path)
 assert exists(weights_path)
@@ -171,10 +175,7 @@ for chunk in ut.ichunks(dataloader, BATCH_SIZE):
         size = img.shape[:2][::-1]
 
         img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
-        img = ln.data.transform.Letterbox.apply(
-            img,
-            dimension=INPUT_SIZE
-        )
+        img = ln.data.transform.Letterbox.apply(img, dimension=INPUT_SIZE)
         img = transform(img)
 
         inputs_.append(img)
@@ -213,11 +214,7 @@ best_confusion = None
 for thresh in tqdm.tqdm(threshs):
     globals().update(locals())
     values = [
-        [
-            prediction
-            for prediction in predictions
-            if prediction.confidence >= thresh
-        ]
+        [prediction for prediction in predictions if prediction.confidence >= thresh]
         for predictions in predictions_pytorch
     ]
     values = [len(value) > 0 for value in values]
@@ -236,14 +233,19 @@ print(f'TN:        {tn}')
 print(f'FP:        {fp}')
 print(f'FN:        {fn}')
 
-"""
-Thresh:    0.48
-Accuracy:  0.965
-TP:        94
-TN:        99
-FP:        1
-FN:        6
-"""
+# Thresh:    0.25
+# Accuracy:  0.93
+# TP:        88
+# TN:        98
+# FP:        2
+# FN:        12
+
+# Thresh:    0.35
+# Accuracy:  0.925
+# TP:        85
+# TN:        100
+# FP:        0
+# FN:        15
 
 #############
 
@@ -259,7 +261,11 @@ output = torch.onnx.export(
     onnx_filename,
     verbose=True,
     input_names=input_names,
-    output_names=output_names
+    output_names=output_names,
+    dynamic_axes={
+        'input': {0: 'batch_size'},  # variable length axes
+        'output': {0: 'batch_size'},
+    },
 )
 
 ###########
@@ -276,14 +282,12 @@ num_classes = params.network.num_classes
 anchors = params.network.anchors
 network_size = (INPUT_SIZE_H, INPUT_SIZE_W, 3)
 class_label_map = params.class_label_map
-conf_thresh = 0.4
-nms_thresh = 1.0
+conf_thresh = 0.0
+nms_thresh = 0.2
 
 postprocess = ln.data.transform.Compose(
     [
-        ln.data.transform.GetBoundingBoxes(
-            num_classes, anchors, conf_thresh
-        ),
+        ln.data.transform.GetBoundingBoxes(num_classes, anchors, conf_thresh),
         ln.data.transform.NonMaxSupression(nms_thresh),
         ln.data.transform.TensorToBrambox(network_size, class_label_map),
     ]
@@ -299,7 +303,7 @@ for chunk in ut.ichunks(zipped, BATCH_SIZE):
     sizes_ = ut.take_column(chunk, 1)
 
     trim = len(imgs)
-    while(len(imgs)) < BATCH_SIZE:
+    while (len(imgs)) < BATCH_SIZE:
         imgs.append(np.random.randn(3, INPUT_SIZE_H, INPUT_SIZE_W).astype(np.float32))
         sizes_.append(INPUT_SIZE)
     input_ = np.array(imgs, dtype=np.float32)
@@ -328,19 +332,11 @@ predictions_onnx = outputs
 
 globals().update(locals())
 values_pytorch = [
-    [
-        prediction
-        for prediction in predictions
-        if prediction.confidence >= best_thresh
-    ]
+    [prediction for prediction in predictions if prediction.confidence >= best_thresh]
     for predictions in predictions_pytorch
 ]
 values_onnx = [
-    [
-        prediction
-        for prediction in predictions
-        if prediction.confidence >= best_thresh
-    ]
+    [prediction for prediction in predictions if prediction.confidence >= best_thresh]
     for predictions in predictions_onnx
 ]
 
@@ -350,9 +346,7 @@ for value_pytorch, value_onnx in zip(values_pytorch, values_onnx):
     for value_p, value_o in zip(value_pytorch, value_onnx):
         assert value_p.class_label == value_o.class_label
         for attr in ['x_top_left', 'y_top_left', 'width', 'height', 'confidence']:
-            deviation = abs(
-                getattr(value_p, attr) - getattr(value_o, attr)
-            )
+            deviation = abs(getattr(value_p, attr) - getattr(value_o, attr))
             deviations.append(deviation)
 
 print(f'Min:  {np.min(deviations):0.08f}')
@@ -362,11 +356,7 @@ print(f'Time Pytorch: {time_pytorch:0.02f} sec.')
 print(f'Time ONNX:    {time_onnx:0.02f} sec.')
 
 values = [
-    [
-        prediction
-        for prediction in predictions
-        if prediction.confidence >= best_thresh
-    ]
+    [prediction for prediction in predictions if prediction.confidence >= best_thresh]
     for predictions in predictions_onnx
 ]
 values = [len(value) > 0 for value in values]
@@ -374,21 +364,33 @@ accuracy = sklearn.metrics.accuracy_score(targets, values)
 confusion = sklearn.metrics.confusion_matrix(targets, values)
 tn, fp, fn, tp = best_confusion.ravel()
 
+print(f'Thresh:    {best_thresh}')
 print(f'Accuracy:  {best_accuracy}')
 print(f'TP:        {tp}')
 print(f'TN:        {tn}')
 print(f'FP:        {fp}')
 print(f'FN:        {fn}')
 
-"""
-Min:  0.00000000
-Max:  0.00012207
-Mean: 0.00001001 +/- 0.00001388
-Time Pytorch: 18.31 sec.
-Time ONNX:    10.37 sec.
-Accuracy:  0.965
-TP:        94
-TN:        99
-FP:        1
-FN:        6
-"""
+# Min:  0.00000000
+# Max:  0.00017841
+# Mean: 0.00000904 +/- 0.00001550
+# Time Pytorch: 18.18 sec.
+# Time ONNX:    9.77 sec.
+# Thresh:    0.25
+# Accuracy:  0.93
+# TP:        88
+# TN:        98
+# FP:        2
+# FN:        12
+
+# Min:  0.00000000
+# Max:  0.00011268
+# Mean: 0.00000845 +/- 0.00001284
+# Time Pytorch: 18.75 sec.
+# Time ONNX:    9.72 sec.
+# Thresh:    0.35000000000000003
+# Accuracy:  0.925
+# TP:        85
+# TN:        100
+# FP:        0
+# FN:        15

scoutbot/loc/models/onnx/scout.loc.5fbfff26.0.onnx

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5f186ab6455efb49769391b678a21cdad7243758bc9486b9bdbab1ab0ed7d611
-size 202393042
+oid sha256:85a9378311d42b5143f74570136f32f50bf97c548135921b178b46ba7612b216
+size 202392948

scoutbot/loc/models/onnx/scout.loc.5fbfff26.1.onnx

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5afe32a0e442ef356ad9eec5f8558ac9fe0cc7edca0695a8ba299995d978143a
-size 202393042
+oid sha256:1f4b48d2afb65a601372b9ed6294a3e93bfdf4a096cce2bc43a6bf857c5029f9
+size 202392948

scoutbot/loc/models/pytorch/detect.lightnet.scout.5fbfff26.v0.py

@@ -1,3 +1,4 @@
+# -*- coding: utf-8 -*-
 import lightnet as ln
 import torch
 
@@ -6,47 +7,51 @@ __all__ = ['params']
 
 params = ln.engine.HyperParameters(
     # Network
-    class_label_map = ['elephant_savanna'],
-    input_dimension = (416, 416),
-    batch_size = 64,
-    mini_batch_size = 8,
-    max_batches = 60000,
-
+    class_label_map=['elephant_savanna'],
+    input_dimension=(416, 416),
+    batch_size=64,
+    mini_batch_size=8,
+    max_batches=60000,
     # Dataset
-    _train_set = '/data/ibeis/ELPH_Vulcan_Final/_ibsdb/_ibeis_cache/training/lightnet/lightnet-training-elephant_savanna-1089e6c2d3b95283/data/train.pkl',
-    _valid_set = None,
-    _test_set = '/data/ibeis/ELPH_Vulcan_Final/_ibsdb/_ibeis_cache/training/lightnet/lightnet-training-elephant_savanna-1089e6c2d3b95283/data/test.pkl',
-    _filter_anno = 'ignore',
-
+    _train_set='/data/ibeis/ELPH_Vulcan_Final/_ibsdb/_ibeis_cache/training/lightnet/lightnet-training-elephant_savanna-1089e6c2d3b95283/data/train.pkl',
+    _valid_set=None,
+    _test_set='/data/ibeis/ELPH_Vulcan_Final/_ibsdb/_ibeis_cache/training/lightnet/lightnet-training-elephant_savanna-1089e6c2d3b95283/data/test.pkl',
+    _filter_anno='ignore',
     # Data Augmentation
-    jitter = .3,
-    flip = .5,
-    hue = .1,
-    saturation = 1.5,
-    value = 1.5,
+    jitter=0.3,
+    flip=0.5,
+    hue=0.1,
+    saturation=1.5,
+    value=1.5,
 )
 
+
 # Network
 def init_weights(m):
     if isinstance(m, torch.nn.Conv2d):
         torch.nn.init.kaiming_normal_(m.weight, nonlinearity='leaky_relu')
 
+
 params.network = ln.models.Yolo(
     len(params.class_label_map),
-    conf_thresh = .001,
-    nms_thresh = .5,
+    conf_thresh=0.001,
+    nms_thresh=0.5,
+)
+params.network.postprocess.append(
+    ln.data.transform.TensorToBrambox(params.input_dimension, params.class_label_map)
 )
-params.network.postprocess.append(ln.data.transform.TensorToBrambox(params.input_dimension, params.class_label_map))
 params.network.apply(init_weights)
 
 # Optimizers
-params.add_optimizer(torch.optim.SGD(
-    params.network.parameters(),
-    lr = .001 / params.batch_size,
-    momentum = .9,
-    weight_decay = .0005 * params.batch_size,
-    dampening = 0,
-))
+params.add_optimizer(
+    torch.optim.SGD(
+        params.network.parameters(),
+        lr=0.001 / params.batch_size,
+        momentum=0.9,
+        weight_decay=0.0005 * params.batch_size,
+        dampening=0,
+    )
+)
 
 # Schedulers
 burn_in = torch.optim.lr_scheduler.LambdaLR(
@@ -55,11 +60,13 @@ burn_in = torch.optim.lr_scheduler.LambdaLR(
 )
 step = torch.optim.lr_scheduler.MultiStepLR(
     params.optimizers[0],
-    milestones = [20000, 40000],
-    gamma = .1,
+    milestones=[20000, 40000],
+    gamma=0.1,
+)
+params.add_scheduler(
+    ln.engine.SchedulerCompositor(
+        #   batch   scheduler
+        (0, burn_in),
+        (1000, step),
+    )
 )
-params.add_scheduler(ln.engine.SchedulerCompositor(
-#   batch   scheduler
-    (0,     burn_in),
-    (1000,  step),
-))

scoutbot/loc/models/pytorch/detect.lightnet.scout.5fbfff26.v1.py

@@ -1,3 +1,4 @@
+# -*- coding: utf-8 -*-
 import lightnet as ln
 import torch
 
@@ -6,47 +7,51 @@ __all__ = ['params']
 
 params = ln.engine.HyperParameters(
     # Network
-    class_label_map = ['elephant_savanna'],
-    input_dimension = (416, 416),
-    batch_size = 64,
-    mini_batch_size = 8,
-    max_batches = 60000,
-
+    class_label_map=['elephant_savanna'],
+    input_dimension=(416, 416),
+    batch_size=64,
+    mini_batch_size=8,
+    max_batches=60000,
     # Dataset
-    _train_set = '/data/ibeis/ELPH_Vulcan_Final/_ibsdb/_ibeis_cache/training/lightnet/lightnet-training-elephant_savanna-c7352b1d409e865d/data/train.pkl',
-    _valid_set = None,
-    _test_set = '/data/ibeis/ELPH_Vulcan_Final/_ibsdb/_ibeis_cache/training/lightnet/lightnet-training-elephant_savanna-c7352b1d409e865d/data/test.pkl',
-    _filter_anno = 'ignore',
-
+    _train_set='/data/ibeis/ELPH_Vulcan_Final/_ibsdb/_ibeis_cache/training/lightnet/lightnet-training-elephant_savanna-c7352b1d409e865d/data/train.pkl',
+    _valid_set=None,
+    _test_set='/data/ibeis/ELPH_Vulcan_Final/_ibsdb/_ibeis_cache/training/lightnet/lightnet-training-elephant_savanna-c7352b1d409e865d/data/test.pkl',
+    _filter_anno='ignore',
     # Data Augmentation
-    jitter = .3,
-    flip = .5,
-    hue = .1,
-    saturation = 1.5,
-    value = 1.5,
+    jitter=0.3,
+    flip=0.5,
+    hue=0.1,
+    saturation=1.5,
+    value=1.5,
 )
 
+
 # Network
 def init_weights(m):
     if isinstance(m, torch.nn.Conv2d):
         torch.nn.init.kaiming_normal_(m.weight, nonlinearity='leaky_relu')
 
+
 params.network = ln.models.Yolo(
     len(params.class_label_map),
-    conf_thresh = .001,
-    nms_thresh = .5,
+    conf_thresh=0.001,
+    nms_thresh=0.5,
+)
+params.network.postprocess.append(
+    ln.data.transform.TensorToBrambox(params.input_dimension, params.class_label_map)
 )
-params.network.postprocess.append(ln.data.transform.TensorToBrambox(params.input_dimension, params.class_label_map))
 params.network.apply(init_weights)
 
 # Optimizers
-params.add_optimizer(torch.optim.SGD(
-    params.network.parameters(),
-    lr = .001 / params.batch_size,
-    momentum = .9,
-    weight_decay = .0005 * params.batch_size,
-    dampening = 0,
-))
+params.add_optimizer(
+    torch.optim.SGD(
+        params.network.parameters(),
+        lr=0.001 / params.batch_size,
+        momentum=0.9,
+        weight_decay=0.0005 * params.batch_size,
+        dampening=0,
+    )
+)
 
 # Schedulers
 burn_in = torch.optim.lr_scheduler.LambdaLR(
@@ -55,11 +60,13 @@ burn_in = torch.optim.lr_scheduler.LambdaLR(
 )
 step = torch.optim.lr_scheduler.MultiStepLR(
     params.optimizers[0],
-    milestones = [20000, 40000],
-    gamma = .1,
+    milestones=[20000, 40000],
+    gamma=0.1,
+)
+params.add_scheduler(
+    ln.engine.SchedulerCompositor(
+        #   batch   scheduler
+        (0, burn_in),
+        (1000, step),
+    )
 )
-params.add_scheduler(ln.engine.SchedulerCompositor(
-#   batch   scheduler
-    (0,     burn_in),
-    (1000,  step),
-))

scoutbot/loc/transforms/__init__.py

@@ -4,6 +4,11 @@
 #   Copyright EAVISE
 #
 
-from ._preprocess import *
-from ._postprocess import *
-from .util import *
+from ._postprocess import (  # NOQA
+    GetBoundingBoxes,
+    NonMaxSupression,
+    ReverseLetterbox,
+    TensorToBrambox,
+)
+from ._preprocess import Letterbox  # NOQA
+from .util import Compose  # NOQA

scoutbot/loc/transforms/_postprocess.py

@@ -6,9 +6,12 @@
 #
 
 import logging
+
 import torch
+
 # from torch.autograd import Variable
 from scoutbot.loc.transforms.detections.detection import Detection
+
 from .util import BaseTransform
 
 __all__ = [
@@ -21,7 +24,7 @@ log = logging.getLogger(__name__)
 
 
 class GetBoundingBoxes(BaseTransform):
-    """ Convert output from darknet networks to bounding box tensor.
+    """Convert output from darknet networks to bounding box tensor.
 
     Args:
         num_classes (int): number of categories
@@ -119,7 +122,7 @@ class GetBoundingBoxes(BaseTransform):
 
 
 class NonMaxSupression(BaseTransform):
-    """ Performs nms on the bounding boxes, filtering boxes with a high overlap.
+    """Performs nms on the bounding boxes, filtering boxes with a high overlap.
 
     Args:
         nms_thresh (Number [0-1]): Overlapping threshold to filter detections with non-maxima suppresion
@@ -142,7 +145,7 @@ class NonMaxSupression(BaseTransform):
 
     @staticmethod
     def _nms(boxes, nms_thresh, class_nms):
-        """ Non maximum suppression.
+        """Non maximum suppression.
 
         Args:
           boxes (tensor): Bounding boxes of one image
@@ -182,7 +185,7 @@ class NonMaxSupression(BaseTransform):
             conflicting = conflicting & same_class
 
         conflicting = conflicting.cpu()
-        keep = torch.zeros(len(conflicting), dtype=torch.uint8)
+        keep = torch.zeros(len(conflicting), dtype=bool)
         supress = torch.zeros(len(conflicting), dtype=torch.float)
         for i, row in enumerate(conflicting):
             if not supress[i]:
@@ -193,7 +196,7 @@ class NonMaxSupression(BaseTransform):
 
 
 class TensorToBrambox(BaseTransform):
-    """ Converts a tensor to a list of brambox objects.
+    """Converts a tensor to a list of brambox objects.
 
     Args:
         network_size (tuple): Tuple containing the width and height of the images going in the network
@@ -255,7 +258,7 @@ class TensorToBrambox(BaseTransform):
 
 
 class ReverseLetterbox(BaseTransform):
-    """ Performs a reverse letterbox operation on the bounding boxes, so they can be visualised on the original image.
+    """Performs a reverse letterbox operation on the bounding boxes, so they can be visualised on the original image.
 
     Args:
         network_size (tuple): Tuple containing the width and height of the images going in the network

scoutbot/loc/transforms/_preprocess.py

@@ -7,8 +7,10 @@
 #
 import collections
 import logging
+
 import numpy as np
 from PIL import Image, ImageOps
+
 from .util import BaseMultiTransform
 
 log = logging.getLogger(__name__)
@@ -23,7 +25,7 @@ __all__ = ['Letterbox']
 
 
 class Letterbox(BaseMultiTransform):
-    """ Transform images and annotations to the right network dimensions.
+    """Transform images and annotations to the right network dimensions.
 
     Args:
         dimension (tuple, optional): Default size for the letterboxing, expressed as a (width, height) tuple; Default **None**
@@ -61,7 +63,7 @@ class Letterbox(BaseMultiTransform):
             return data
 
     def _tf_pil(self, img):
-        """ Letterbox an image to fit in the network """
+        """Letterbox an image to fit in the network"""
         if self.dataset is not None:
             net_w, net_h = self.dataset.input_dim
         else:
@@ -100,7 +102,7 @@ class Letterbox(BaseMultiTransform):
         return img
 
     def _tf_cv(self, img):
-        """ Letterbox and image to fit in the network """
+        """Letterbox and image to fit in the network"""
         if self.dataset is not None:
             net_w, net_h = self.dataset.input_dim
         else:
@@ -144,7 +146,7 @@ class Letterbox(BaseMultiTransform):
         return img
 
     def _tf_anno(self, annos):
-        """ Change coordinates of an annotation, according to the previous letterboxing """
+        """Change coordinates of an annotation, according to the previous letterboxing"""
         for anno in annos:
             if self.scale is not None:
                 anno.x_top_left *= self.scale

scoutbot/loc/transforms/annotations/annotation.py

@@ -12,7 +12,7 @@ __all__ = ['Annotation', 'ParserType', 'Parser']
 
 
 class Annotation(b.Box):
-    """ This is a generic annotation class that provides some common functionality all annotations need.
+    """This is a generic annotation class that provides some common functionality all annotations need.
     It builds upon :class:`~brambox.boxes.box.Box`.
 
     Attributes:
@@ -37,7 +37,7 @@ class Annotation(b.Box):
     """
 
     def __init__(self):
-        """ x_top_left,y_top_left,width,height are in pixel coordinates """
+        """x_top_left,y_top_left,width,height are in pixel coordinates"""
         super(Annotation, self).__init__()
         self.lost = False  # if object is not seen in the image, if true one must ignore this annotation
         self.difficult = False  # if the object is considered difficult
@@ -75,7 +75,7 @@ class Annotation(b.Box):
 
     @classmethod
     def create(cls, obj=None):
-        """ Create an annotation from a string or other box object.
+        """Create an annotation from a string or other box object.
 
         Args:
             obj (Box or string, optional): Bounding box object to copy attributes from or string to deserialize
@@ -113,7 +113,7 @@ class Annotation(b.Box):
         return instance
 
     def __repr__(self):
-        """ Unambiguous representation """
+        """Unambiguous representation"""
         string = f'{self.__class__.__name__} ' + '{'
         string += f"class_label = '{self.class_label}', "
         string += f'object_id = {self.object_id}, '
@@ -134,7 +134,7 @@ class Annotation(b.Box):
         return string + '}'
 
     def __str__(self):
-        """ Pretty print """
+        """Pretty print"""
         string = 'Annotation {'
         string += f'\'{self.class_label}\'{"" if self.object_id is None else " "+str(self.object_id)}, '
         string += f'[{int(self.x_top_left)}, {int(self.y_top_left)}, {int(self.width)}, {int(self.height)}]'
@@ -160,6 +160,6 @@ ParserType = b.ParserType
 
 
 class Parser(b.Parser):
-    """ Generic parser class """
+    """Generic parser class"""
 
     box_type = Annotation  # Derived classes should set the correct box_type

scoutbot/loc/transforms/box.py

@@ -9,7 +9,7 @@ __all__ = ['Box', 'ParserType', 'Parser']
 
 
 class Box:
-    """ This is a generic bounding box representation.
+    """This is a generic bounding box representation.
     This class provides some base functionality to both annotations and detections.
 
     Attributes:
@@ -31,7 +31,7 @@ class Box:
 
     @classmethod
     def create(cls, obj=None):
-        """ Create a bounding box from a string or other detection object.
+        """Create a bounding box from a string or other detection object.
 
         Args:
             obj (Box or string, optional): Bounding box object to copy attributes from or string to deserialize
@@ -62,16 +62,16 @@ class Box:
         return self.__dict__ == other.__dict__
 
     def serialize(self):
-        """ abstract serializer, implement in derived classes. """
+        """abstract serializer, implement in derived classes."""
         raise NotImplementedError
 
     def deserialize(self, string):
-        """ abstract parser, implement in derived classes. """
+        """abstract parser, implement in derived classes."""
         raise NotImplementedError
 
 
 class ParserType(Enum):
-    """ Enum for differentiating between different parser types. """
+    """Enum for differentiating between different parser types."""
 
     UNDEFINED = 0  #: Undefined parsertype. Do not use this!
     SINGLE_FILE = 1  #: One single file contains all annotations
@@ -79,7 +79,7 @@ class ParserType(Enum):
 
 
 class Parser:
-    """ This is a Generic parser class.
+    """This is a Generic parser class.
 
     Args:
         kwargs (optional): Derived parsers should use keyword arguments to get any information they need upon initialisation.
@@ -97,7 +97,7 @@ class Parser:
         pass
 
     def serialize(self, box):
-        """ Serialization function that can be overloaded in the derived class.
+        """Serialization function that can be overloaded in the derived class.
         The default serializer will call the serialize function of the bounding boxes and join them with a newline.
 
         Args:
@@ -124,7 +124,7 @@ class Parser:
         return result
 
     def deserialize(self, string):
-        """ Deserialization function that can be overloaded in the derived class.
+        """Deserialization function that can be overloaded in the derived class.
         The default deserialize will create new ``box_type`` objects and call the deserialize function of these objects with every line of the input string.
 
         Args:

scoutbot/loc/transforms/detections/detection.py

@@ -12,7 +12,7 @@ __all__ = ['Detection', 'ParserType', 'Parser']
 
 
 class Detection(b.Box):
-    """ This is a generic detection class that provides some base functionality all detections need.
+    """This is a generic detection class that provides some base functionality all detections need.
     It builds upon :class:`~brambox.boxes.box.Box`.
 
     Attributes:
@@ -20,13 +20,13 @@ class Detection(b.Box):
     """
 
     def __init__(self):
-        """ x_top_left,y_top_left,width,height are in pixel coordinates """
+        """x_top_left,y_top_left,width,height are in pixel coordinates"""
         super(Detection, self).__init__()
         self.confidence = 0.0  # Confidence score between 0-1
 
     @classmethod
     def create(cls, obj=None):
-        """ Create a detection from a string or other box object.
+        """Create a detection from a string or other box object.
 
         Args:
             obj (Box or string, optional): Bounding box object to copy attributes from or string to deserialize
@@ -48,7 +48,7 @@ class Detection(b.Box):
         return instance
 
     def __repr__(self):
-        """ Unambiguous representation """
+        """Unambiguous representation"""
         string = f'{self.__class__.__name__} ' + '{'
         string += f'class_label = {self.class_label}, '
         string += f'object_id = {self.object_id}, '
@@ -60,7 +60,7 @@ class Detection(b.Box):
         return string + '}'
 
     def __str__(self):
-        """ Pretty print """
+        """Pretty print"""
         string = 'Detection {'
         string += f'\'{self.class_label}\'{"" if self.object_id is None else " "+str(self.object_id)}, '
         string += f'[{int(self.x_top_left)}, {int(self.y_top_left)}, {int(self.width)}, {int(self.height)}]'
@@ -107,6 +107,6 @@ ParserType = b.ParserType
 
 
 class Parser(b.Parser):
-    """ Generic parser class """
+    """Generic parser class"""
 
     box_type = Detection  # Derived classes should set the correct box_type

scoutbot/loc/transforms/util.py

@@ -11,7 +11,7 @@ __all__ = ['Compose']
 
 
 class Compose(list):
-    """ This is lightnet's own version of :class:`torchvision.transforms.Compose`.
+    """This is lightnet's own version of :class:`torchvision.transforms.Compose`.
 
     Note:
         The reason we have our own version is because this one offers more freedom to the user.
@@ -55,7 +55,7 @@ class Compose(list):
 
 
 class BaseTransform(ABC):
-    """ Base transform class for the pre- and post-processing functions.
+    """Base transform class for the pre- and post-processing functions.
     This class allows to create an object with some case specific settings, and then call it with the data to perform the transformation.
     It also allows to call the static method ``apply`` with the data and settings. This is usefull if you want to transform a single data object.
     """
@@ -70,7 +70,7 @@ class BaseTransform(ABC):
     @classmethod
     @abstractmethod
     def apply(cls, data, **kwargs):
-        """ Classmethod that applies the transformation once.
+        """Classmethod that applies the transformation once.
 
         Args:
             data: Data to transform (eg. image)
@@ -80,7 +80,7 @@ class BaseTransform(ABC):
 
 
 class BaseMultiTransform(ABC):
-    """ Base multiple transform class that is mainly used in pre-processing functions.
+    """Base multiple transform class that is mainly used in pre-processing functions.
     This class exists for transforms that affect both images and annotations.
     It provides a classmethod ``apply``, that will perform the transormation on one (data, target) pair.
     """
@@ -95,7 +95,7 @@ class BaseMultiTransform(ABC):
 
     @classmethod
     def apply(cls, data, target=None, **kwargs):
-        """ Classmethod that applies the transformation once.
+        """Classmethod that applies the transformation once.
 
         Args:
             data: Data to transform (eg. image)

scoutbot/scoutbot.py

@@ -5,15 +5,17 @@ The lecture materials for Lecture 1: Dataset Prototyping and Visualization
 """
 import click
 
+from scoutbot import utils
+
+log = utils.init_logging()
+
 
 @click.command()
 @click.option(
     '--config', help='Path to config file', default='configs/mnist_resnet18.yaml'
 )
 def wic(config):
-    """
-    
-    """
+    """ """
     pass
 
 
@@ -22,9 +24,7 @@ def wic(config):
     '--config', help='Path to config file', default='configs/mnist_resnet18.yaml'
 )
 def main(config):
-    """
-    
-    """
+    """ """
     pass
 
 

scoutbot/tile/__init__.py

@@ -2,6 +2,146 @@
 '''
 2022 Wild Me
 '''
+from os.path import abspath, exists, join, split, splitext
 
-version = '0.1.0'
-__version__ = version
+import cv2
+import numpy as np
+
+TILE_WIDTH = 256
+TILE_HEIGHT = 256
+TILE_SIZE = (TILE_WIDTH, TILE_HEIGHT)
+TILE_OVERLAP = 64
+TILE_OFFSET = 0
+TILE_BORDERS = True
+
+
+def compute(img_filepath, grid1=True, grid2=True, ext=None, **kwargs):
+    """Compute the tiles for a given input image"""
+    assert exists(img_filepath)
+    img = cv2.imread(img_filepath)
+
+    grids = []
+    if grid1:
+        grids += tile_grid(img.shape)
+    if grid2:
+        grids += tile_grid(img.shape, offset=TILE_WIDTH // 2, borders=False)
+
+    filepaths = [tile_filepath(img_filepath, grid, ext=ext) for grid in grids]
+    for grid, filepath in zip(grids, filepaths):
+        assert tile_write(img, grid, filepath)
+
+    return filepaths
+
+
+def tile_write(img, grid, filepath):
+    if exists(filepath):
+        return True
+
+    x0 = grid.get('x')
+    y0 = grid.get('y')
+    w = grid.get('w')
+    h = grid.get('h')
+    y1 = y0 + h
+    x1 = x0 + w
+
+    tile = img[y0:y1, x0:x1]
+    cv2.imwrite(filepath, tile)
+    return exists(filepath)
+
+
+def tile_filepath(img_filepath, grid, ext=None):
+    x = grid.get('x')
+    y = grid.get('y')
+    w = grid.get('w')
+    h = grid.get('h')
+
+    assert exists(img_filepath)
+    img_filepath = abspath(img_filepath)
+
+    img_path, img_filename = split(img_filepath)
+    img_name, img_ext = splitext(img_filename)
+
+    img_ext = img_ext if ext is None else ext
+
+    filepath = join(img_path, f'{img_name}_x_{x}_y_{y}_w_{w}_h_{h}{img_ext}')
+    return filepath
+
+
+def tile_grid(
+    shape, size=TILE_SIZE, overlap=TILE_OVERLAP, offset=TILE_OFFSET, borders=TILE_BORDERS
+):
+    h_, w_ = shape[:2]
+    w, h = size
+    ol = overlap
+    os = offset
+
+    if borders:
+        assert offset == 0, 'Cannot use an offset with borders turned on'
+
+    y_ = int(np.floor((h_ - ol) / (h - ol)))
+    x_ = int(np.floor((w_ - ol) / (w - ol)))
+    iy = (h * y_) - (ol * (y_ - 1))
+    ix = (w * x_) - (ol * (x_ - 1))
+    oy = int(np.floor((h_ - iy) * 0.5))
+    ox = int(np.floor((w_ - ix) * 0.5))
+
+    miny = 0
+    minx = 0
+    maxy = h_ - h
+    maxx = w_ - w
+
+    ys = list(range(oy, h_ - h + 1, h - ol))
+    yb = [False] * len(ys)
+    xs = list(range(ox, w_ - w + 1, w - ol))
+    xb = [False] * len(xs)
+
+    if borders and oy > 0:
+        ys = [miny] + ys + [maxy]
+        yb = [True] + yb + [True]
+
+    if borders and ox > 0:
+        xs = [minx] + xs + [maxx]
+        xb = [True] + xb + [True]
+
+    outputs = []
+    for y0, yb_ in zip(ys, yb):
+        y0 += os
+        y1 = y0 + h
+        for x0, xb_ in zip(xs, xb):
+            x0 += os
+            x1 = x0 + w
+
+            # Sanity, mostly to check for offset
+            valid = True
+            try:
+                assert x1 - x0 == w, '%d, %d' % (
+                    x1 - x0,
+                    w,
+                )
+                assert y1 - y0 == h, '%d, %d' % (
+                    y1 - y0,
+                    h,
+                )
+                assert 0 <= x0 and x0 <= w_, '%d, %d' % (
+                    x0,
+                    w_,
+                )
+                assert 0 <= x1 and x1 <= w_, '%d, %d' % (
+                    x1,
+                    w_,
+                )
+                assert 0 <= y0 and y0 <= h_, '%d, %d' % (
+                    y0,
+                    h_,
+                )
+                assert 0 <= y1 and y1 <= h_, '%d, %d' % (
+                    y1,
+                    h_,
+                )
+            except AssertionError:
+                valid = False
+
+            if valid:
+                outputs.append({'x': x0, 'y': y0, 'w': w, 'h': h, 'b': yb_ or xb_})
+
+    return outputs

scoutbot/utils.py

@@ -5,7 +5,6 @@
 import logging
 from logging.handlers import TimedRotatingFileHandler
 
-
 DAYS = 21
 
 

scoutbot/wic/__init__.py

@@ -3,13 +3,19 @@
 2022 Wild Me
 '''
 from os.path import join
-import onnxruntime as ort
 from pathlib import Path
-from scoutbot.wic.dataloader import _init_transforms, ImageFilePathList, BATCH_SIZE, INPUT_SIZE
+
 import numpy as np
-import utool as ut
+import onnxruntime as ort
 import torch
+import utool as ut
 
+from scoutbot.wic.dataloader import (
+    BATCH_SIZE,
+    INPUT_SIZE,
+    ImageFilePathList,
+    _init_transforms,
+)
 
 PWD = Path(__file__).absolute().parent
 
@@ -25,23 +31,23 @@ def pre(inputs):
     )
 
     data = []
-    for data_, in dataloader:
+    for (data_,) in dataloader:
         data += data_.tolist()
 
     return data
 
 
-def predict(data):
-    ort_session = ort.InferenceSession(
-        ONNX_MODEL, 
-        providers=['CPUExecutionProvider']
-    )
+def predict(data, fill=False):
+    ort_session = ort.InferenceSession(ONNX_MODEL, providers=['CPUExecutionProvider'])
 
     preds = []
     for chunk in ut.ichunks(data, BATCH_SIZE):
         trim = len(chunk)
-        while(len(chunk)) < BATCH_SIZE:
-            chunk.append(np.random.randn(3, INPUT_SIZE, INPUT_SIZE).astype(np.float32))
+        if fill:
+            while (len(chunk)) < BATCH_SIZE:
+                chunk.append(
+                    np.random.randn(3, INPUT_SIZE, INPUT_SIZE).astype(np.float32)
+                )
         input_ = np.array(chunk, dtype=np.float32)
 
         pred_ = ort_session.run(
@@ -54,8 +60,5 @@ def predict(data):
 
 
 def post(preds):
-    outputs = [
-        dict(zip(ONNX_CLASSES, pred)) 
-        for pred in preds
-    ]
+    outputs = [dict(zip(ONNX_CLASSES, pred)) for pred in preds]
     return outputs

scoutbot/wic/convert.py

@@ -1,32 +1,37 @@
+# -*- coding: utf-8 -*-
 """
 
 pip install torch torchvision onnx onnxruntime-gpu tqdm wbia-utool scikit-learn numpy
 
 """
+import random
+import time
+from collections import OrderedDict
+from os.path import exists, join, split, splitext
+
+import numpy as np
+import onnx
+import onnxruntime as ort
+import sklearn
 import torch
 import torch.nn as nn
 import torchvision
-import onnx
-import onnxruntime as ort
 import tqdm
-import random
 import utool as ut
-import numpy as np
-import sklearn
-import time
-from os.path import join, exists, split, splitext
-from wbia.algo.detect.densenet import INPUT_SIZE, _init_transforms, ImageFilePathList
-
+import wbia
+from wbia.algo.detect.densenet import INPUT_SIZE, ImageFilePathList, _init_transforms
 
 WITH_GPU = False
 BATCH_SIZE = 128
 
 
+ibs = wbia.opendb(dbdir='/data/db')
+
+
 pkl_path = 'scout.pkl'
 if not exists(pkl_path):
     if False:
-        pass
-        # tids = ibs.get_valid_gids(is_tile=True)
+        tids = ibs.get_valid_gids(is_tile=True)
     else:
         imageset_text_list = ['TEST_SET']
         imageset_rowid_list = ibs.get_imageset_imgsetids_from_text(imageset_text_list)
@@ -42,7 +47,9 @@ if not exists(pkl_path):
     for chunk_tids in tqdm.tqdm(ut.ichunks(tids, 1000)):
         _, _, chunk_flags = ibs.scout_tile_positive_cumulative_area(chunk_tids)
         chunk_filepaths = ibs.get_image_paths(chunk_tids)
-        for index, (tid, flag, filepath) in enumerate(zip(chunk_tids, chunk_flags, chunk_filepaths)):
+        for index, (tid, flag, filepath) in enumerate(
+            zip(chunk_tids, chunk_flags, chunk_filepaths)
+        ):
             if not exists(filepath):
                 continue
             if flag:
@@ -85,7 +92,7 @@ assert sum(map(exists, filepaths)) == len(filepaths)
 
 ##########
 
-INDEX = 2
+INDEX = 0
 
 weights_path = f'/cache/wbia/classifier2.scout.5fbfff26.3/classifier2.vulcan.5fbfff26.3/classifier.{INDEX}.weights'
 
@@ -100,8 +107,6 @@ num_ftrs = model.classifier.in_features
 model.classifier = nn.Linear(num_ftrs, len(classes))
 
 # Convert any weights to non-parallel version
-from collections import OrderedDict
-
 new_state = OrderedDict()
 for k, v in state.items():
     k = k.replace('module.', '')
@@ -155,10 +160,7 @@ best_accuracy = 0.0
 best_confusion = None
 for thresh in tqdm.tqdm(threshs):
     globals().update(locals())
-    values = [
-        prediction['positive'] >= thresh
-        for prediction in predictions_pytorch
-    ]
+    values = [prediction['positive'] >= thresh for prediction in predictions_pytorch]
     accuracy = sklearn.metrics.accuracy_score(targets, values)
     confusion = sklearn.metrics.confusion_matrix(targets, values)
     if accuracy > best_accuracy:
@@ -174,6 +176,27 @@ print(f'TN:        {tn}')
 print(f'FP:        {fp}')
 print(f'FN:        {fn}')
 
+# Thresh:    0.01
+# Accuracy:  0.895
+# TP:        83
+# TN:        96
+# FP:        4
+# FN:        17
+
+# Thresh:    0.06
+# Accuracy:  0.91
+# TP:        85
+# TN:        97
+# FP:        3
+# FN:        15
+
+# Thresh:    0.01
+# Accuracy:  0.905
+# TP:        83
+# TN:        98
+# FP:        2
+# FN:        17
+
 #############
 
 dummy_input = torch.randn(BATCH_SIZE, 3, INPUT_SIZE, INPUT_SIZE, device='cpu')
@@ -187,7 +210,11 @@ output = torch.onnx.export(
     onnx_filename,
     verbose=True,
     input_names=input_names,
-    output_names=output_names
+    output_names=output_names,
+    dynamic_axes={
+        'input': {0: 'batch_size'},  # variable length axes
+        'output': {0: 'batch_size'},
+    },
 )
 
 ###########
@@ -204,7 +231,7 @@ time_onnx = 0.0
 outputs = []
 for chunk in ut.ichunks(inputs, BATCH_SIZE):
     trim = len(chunk)
-    while(len(chunk)) < BATCH_SIZE:
+    while (len(chunk)) < BATCH_SIZE:
         chunk.append(np.random.randn(3, INPUT_SIZE, INPUT_SIZE).astype(np.float32))
     input_ = np.array(chunk, dtype=np.float32)
 
@@ -222,7 +249,9 @@ predictions_onnx = [dict(zip(classes, output)) for output in outputs]
 
 ###########
 
-values_pytorch = [prediction_pytorch['positive'] for prediction_pytorch in predictions_pytorch]
+values_pytorch = [
+    prediction_pytorch['positive'] for prediction_pytorch in predictions_pytorch
+]
 values_onnx = [prediction_onnx['positive'] for prediction_onnx in predictions_onnx]
 deviations = [
     abs(value_pytorch - value_onnx)
@@ -236,29 +265,51 @@ print(f'Time Pytorch: {time_pytorch:0.02f} sec.')
 print(f'Time ONNX:    {time_onnx:0.02f} sec.')
 
 globals().update(locals())
-values = [
-    prediction['positive'] >= best_thresh
-    for prediction in predictions_onnx
-]
+values = [prediction['positive'] >= best_thresh for prediction in predictions_onnx]
 accuracy = sklearn.metrics.accuracy_score(targets, values)
 confusion = sklearn.metrics.confusion_matrix(targets, values)
 tn, fp, fn, tp = best_confusion.ravel()
 
+print(f'Thresh:    {best_thresh}')
 print(f'Accuracy:  {best_accuracy}')
 print(f'TP:        {tp}')
 print(f'TN:        {tn}')
 print(f'FP:        {fp}')
 print(f'FN:        {fn}')
 
-"""
-Min:  0.00000000
-Max:  0.00000256
-Mean: 0.00000003 +/- 0.00000021
-Time Pytorch: 10.78 sec.
-Time ONNX:    3.83 sec.
-Accuracy:  0.99
-TP:        98
-TN:        100
-FP:        0
-FN:        2
-"""
+# Min:  0.00000000
+# Max:  0.00000143
+# Mean: 0.00000003 +/- 0.00000013
+# Time Pytorch: 9.64 sec.
+# Time ONNX:    3.17 sec.
+# Thresh:    0.01
+# Accuracy:  0.895
+# TP:        83
+# TN:        96
+# FP:        4
+# FN:        17
+
+# Min:  0.00000000
+# Max:  0.00000113
+# Mean: 0.00000004 +/- 0.00000013
+# Time Pytorch: 9.42 sec.
+# Time ONNX:    3.54 sec.
+# Thresh:    0.06
+# Accuracy:  0.91
+# TP:        85
+# TN:        97
+# FP:        3
+# FN:        15
+
+
+# Min:  0.00000000
+# Max:  0.00000209
+# Mean: 0.00000004 +/- 0.00000019
+# Time Pytorch: 9.98 sec.
+# Time ONNX:    3.45 sec.
+# Thresh:    0.01
+# Accuracy:  0.905
+# TP:        83
+# TN:        98
+# FP:        2
+# FN:        17

scoutbot/wic/dataloader.py

@@ -1,11 +1,11 @@
+# -*- coding: utf-8 -*-
+import numpy as np
+import PIL
 import torch
 import torchvision
 import utool as ut
-import numpy as np
-import PIL
-
 
-BATCH_SIZE = 128
+BATCH_SIZE = 2048
 INPUT_SIZE = 224
 
 
@@ -82,7 +82,7 @@ class TestAugmentations(Augmentations):
     def __init__(self, **kwargs):
         from imgaug import augmenters as iaa
 
-        self.aug = iaa.Sequential([iaa.Scale((INPUT_SIZE, INPUT_SIZE))])
+        self.aug = iaa.Sequential([iaa.Resize((INPUT_SIZE, INPUT_SIZE))])
 
 
 def _init_transforms(**kwargs):

scoutbot/wic/models/onnx/scout.wic.5fbfff26.3.0.onnx

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2d9d0d347ca67e770c60f457ac0d85fc592914b6549f93587bdaff22e04fe6e2
-size 75554719
+oid sha256:cbc7f381fa58504e03b6510245b6b2742d63049429337465d95663a6468df4c1
+size 75554737

scoutbot/wic/models/onnx/scout.wic.5fbfff26.3.1.onnx

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5c5689ffb194e0bcd343ee863e317e6a92311850c60c397d1ccff70ce2cc0d20
-size 75555339
+oid sha256:b9779cbfc0d690842abab48403e811b60c5f22f5ebb2ded6db7990ac7d637b33
+size 75554117

scoutbot/wic/models/onnx/scout.wic.5fbfff26.3.2.onnx

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:252b23ea50b4303e66053d533fcce491d9fd2061e5dfd3b0cc0b67efca8d8bdb
-size 75555339
+oid sha256:f3e12bab1391d5b756e854b1bf529ed7107f87c10903094c8fbbf49ed2e57e3c
+size 75554117

setup.cfg

@@ -36,7 +36,7 @@ python_requires = >=3.7
 [options.entry_points]
 console_scripts =
     scoutbot = scoutbot.scoutbot:cli
-    
+
 [bdist_wheel]
 universal = 1
 
@@ -45,10 +45,12 @@ test=pytest
 
 [tool:pytest]
 minversion = 5.4
-addopts = -v -p no:doctest --xdoctest --xdoctest-style=google --random-order --random-order-bucket=global --cov=./ --cov-report html -m "not separate" --durations=0 --durations-min=3.0 --color=yes --code-highlight=yes --show-capture=log -ra
+addopts = -v -p no:doctest --xdoctest --xdoctest-style=google --random-order --random-order-bucket=global --cov=./ --cov-report html -m "not separate" --durations-min=1.0 --color=yes --code-highlight=yes --show-capture=log -ra
 testpaths =
     scoutbot
     tests
+filterwarnings =
+    default
 
 [options.extras_require]
 test =

tests/conftest.py

@@ -1,8 +1,6 @@
 # -*- coding: utf-8 -*-
 import logging
 
-import pytest
-
 log = logging.getLogger('pytest.conftest')  # pylint: disable=invalid-name
 
 

tests/test_loc.py

@@ -1,6 +1,7 @@
 # -*- coding: utf-8 -*-
+from os.path import abspath, exists, join
+
 import onnx
-from os.path import exists, join, abspath
 
 
 def test_loc_onnx_load():
@@ -16,7 +17,7 @@ def test_loc_onnx_load():
 
 
 def test_loc_onnx_pipeline():
-    from scoutbot.loc import pre, predict, post, INPUT_SIZE
+    from scoutbot.loc import INPUT_SIZE, post, pre, predict
 
     inputs = [
         abspath(join('examples', '0d01a14e-311d-e153-356f-8431b6996b84.true.jpg')),
@@ -92,4 +93,4 @@ def test_loc_onnx_pipeline():
             if key == 'class_label':
                 assert getattr(output, key) == target.get(key)
             else:
-                assert abs(getattr(output, key) - target.get(key)) < 1e-6
+                assert abs(getattr(output, key) - target.get(key)) < 1e-2

tests/test_wic.py

@@ -1,6 +1,7 @@
 # -*- coding: utf-8 -*-
+from os.path import abspath, exists, join
+
 import onnx
-from os.path import exists, join, abspath
 
 
 def test_wic_onnx_load():
@@ -16,7 +17,7 @@ def test_wic_onnx_load():
 
 
 def test_wic_onnx_pipeline():
-    from scoutbot.wic import pre, predict, post, ONNX_CLASSES, INPUT_SIZE
+    from scoutbot.wic import INPUT_SIZE, ONNX_CLASSES, post, pre, predict
 
     inputs = [
         abspath(join('examples', '1e8372e4-357d-26e6-d7fd-0e0ae402463a.true.jpg')),
@@ -36,8 +37,8 @@ def test_wic_onnx_pipeline():
     assert len(preds) == 1
     assert len(preds[0]) == 2
     assert preds[0][1] > preds[0][0]
-    assert abs(preds[0][0] - 0.00001503) < 1e-6, str(preds)
-    assert abs(preds[0][1] - 0.99998497) < 1e-6
+    assert abs(preds[0][0] - 0.00001503) < 1e-4
+    assert abs(preds[0][1] - 0.99998497) < 1e-4
 
     outputs = post(preds)
 
@@ -45,5 +46,5 @@ def test_wic_onnx_pipeline():
     output = outputs[0]
     assert output.keys() == set(ONNX_CLASSES)
     assert output['positive'] > output['negative']
-    assert abs(output['negative'] - 0.00001503) < 1e-6
-    assert abs(output['positive'] - 0.99998497) < 1e-6
+    assert abs(output['negative'] - 0.00001503) < 1e-4
+    assert abs(output['positive'] - 0.99998497) < 1e-4