app.py

import gradio as gr
from model import Lightning_YOLO
import config
from utils import non_max_suppression, cells_to_bboxes, draw_bounding_boxes, get_annotations
import torch

scaled_anchors = config.scaled_anchors

model = Lightning_YOLO()
model.load_state_dict(torch.load("yolov3.pth", map_location="cpu"), strict=False)
model.eval()

def inference(image, threst = 0.5, iou_tresh = 0.5):
    image_copy = image.copy()
    transformed_image = config.transforms(image=image)["image"].unsqueeze(0)
    output = model(transformed_image)
    bboxes = [[] for _ in range(1)]
    nms_boxes_output = []
    for i in range(3):
      batch_size, A, S, _, _ = output[i].shape
      anchor = scaled_anchors[i]
      boxes_scale_i = cells_to_bboxes(
          output[i], anchor, S=S, is_preds=True
      )
      for idx, (box) in enumerate(boxes_scale_i):
          bboxes[idx] += box

    
    # nms_boxes = non_max_suppression(
    #   bboxes[0], iou_threshold=iou_tresh, threshold=threst, box_format="midpoint",
    # )
    for i in range(image.shape[0]):
        
        nms_boxes = non_max_suppression(
          bboxes[i], iou_threshold=iou_tresh, threshold=threst, box_format="midpoint",
        )
        nms_boxes_output.append(nms_boxes)
    
    annotations = get_annotations(nms_boxes_output,config.IMAGE_SIZE,config.IMAGE_SIZE)
    # plot_img = draw_bounding_boxes(image.copy(), nms_boxes, class_labels=config.PASCAL_CLASSES)

    # return plot_img
    return [image_copy, annotations]

def visualize_gradcam(image, target_layer=-5, show_cam=True, transparency=0.5):
  # if show_cam:
  #     cam = YoloCAM(model=model, target_layers=[model.layers[target_layer]], use_cuda=False)
  #     transformed_image = config.transforms(image=image)["image"].unsqueeze(0)
  #     grayscale_cam = cam(transformed_image, scaled_anchors)[0, :, :]
  #     img = cv2.resize(image, (416, 416))
  #     img = np.float32(img) / 255
  #     cam_image = show_cam_on_image(img, grayscale_cam, use_rgb=True, image_weight=transparency)
  # else:
  #     cam_image = image
      
  # return cam_image
  pass

window1 = gr.Interface(
    inference,
    inputs=[
        gr.Image(label="Input Image"),
        gr.Slider(0, 1, value=0.5, step=0.1, label="Threshold", info="Set Threshold value"),
        gr.Slider(0, 1, value=0.5, step=0.1, label="IOU Threshold", info="Set IOU Threshold value"),
    ],
    outputs=[
        gr.Image(label="YoloV3 Object Detection"),
    ],
    # examples=ex1,
)


window2 = gr.Interface(
    visualize_gradcam,
    inputs=[
        gr.Image(label="Input Image"),
        gr.Slider(-5, -2, value=-3, step=-1, label="Network Layer", info="GRAD-CAM Layer to visualize?"),
        gr.Checkbox(label="GradCAM", value=True, info="Visualize Class Activation Maps ??"), 
        gr.Slider(0, 1, value=0.5, step=0.1, label="Transparency", info="Set Transparency of GRAD-CAMs"), 
    ],
    outputs=[
        # gr.Image(label="Grad-CAM Visualization"),
        gr.AnnotatedImage(label='BBox Prediction',
                        height=config.IMAGE_SIZE,
                        width=config.IMAGE_SIZE)
    ],
    # examples=ex2,
)


app = gr.TabbedInterface([window1, window2], ["YOLO V3 Detection", "GradCAM Visualization"])
app.launch()