search

varunagrawal / tiny-faces-pytorch

Finding Tiny Faces in PyTorch
MIT License
164 stars 44 forks source link

How can I use the code to test on my own dataset(not WIDER FACE)? #15

Open JackJinWang opened 4 years ago

JackJinWang commented 4 years ago

I use the code to test my own dataset(not wider face).

The parameters is listed as follow: --dataset-root /media/yons/7AD02E63D02E263B1/workplace/facedataset/owndata/Pics2/ /media/yons/7AD02E63D02E263B1/workplace/facedataset/owndata/list2.txt --checkpoint ./checkpoint_50_best.pth --split test

It has some problem as follow:

raceback (most recent call last): File "/media/yons/7AD02E63D02E263B1/workplace/tiny-faces-pytorch/Evaluate.py", line 116, in main() File "/media/yons/7AD02E63D02E263B1/workplace/tiny-faces-pytorch/Evaluate.py", line 112, in main results_dir=args.results_dir, debug=args.debug) File "/media/yons/7AD02E63D02E263B1/workplace/tiny-faces-pytorch/Evaluate.py", line 89, in run nms_thresh, device=device) File "/media/yons/7AD02E63D02E263B1/workplace/tiny-faces-pytorch/trainer.py", line 131, in get_detections templates, prob_thresh, rf, scale) File "/media/yons/7AD02E63D02E263B1/workplace/tiny-faces-pytorch/models/utils.py", line 36, in get_bboxes prob_cls[:, :, invalid_template_id] = 0.0 IndexError: index 22 is out of bounds for axis 2 with size 22

Would you help solve this problem?

JackJinWang commented 4 years ago

The reason may be that the code can not handle images with different size. But I don't know how to solve the problem. Would you please help me?

mengxiangxiang414 commented 2 years ago

Hi,have you solved the problem?

varunagrawal commented 1 month ago

Hi sorry for the delay in responding. It is hard for me to understand what the issue is since I don't have access to your dataset.

My suggestion would be to output the shape of prob_cls. The axis 2 should have size 25 and if it doesn't, you will need to debug why it doesn't.

You can also send me an alternative face dataset that I can try this on.