Chengjian Feng

You can try to clamp the value of the box area when computing GIoU loss, e.g., https://github.com/fcjian/TOOD/blob/93b3a87556e361f7d56507bd56943cf121c3caa2/mmdet/core/bbox/iou_calculators/iou2d_calculator.py#L212-L215

You can move the file of "resnet50-19c8e357.pth" to the path of "~/.cache/torch/hub/checkpoints/".

You are right. It is (n,4). I made a typing error. Thank you for your reminder.

The shorter side of the input image is 800 pixels.

Done, please check README.md.

@jianpingZhonggit tood

You can plot and save the figure with matplotlib, for example: `import matplotlib.pyplot as plt` `fig = plt.figure(figsize=(18,20))` `ax = fig.add_subplot(111)` `im = ax.imshow(score_map, cmap=plt.cm.jet)` `cb = plt.colorbar(im, fraction=0.058)` `cb.ax.tick_params(labelsize=64)`...

@iumyx2612 `score_map ` is the score map predicted from the classifier.

> > @iumyx2612 `score_map ` is the score map predicted from the classifier. > > I was able to produce Classification score map, but how do you get the IoU...

Install PromptDet by running `pip install -v -e .` under the root directory of this project.