search

varunagrawal / tiny-faces-pytorch

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

Could you give some instructions abou how to use it? #3

Closed feixiangdekaka closed 4 years ago

varunagrawal commented 5 years ago

I'm planning to write full docs for this next week. Currently super busy with some deadlines. Sorry and thank you for your patience!

feixiangdekaka commented 5 years ago

I'm planning to write full docs for this next week. Currently super busy with some deadlines. Sorry and thank you for your patience!

I have some bug :+1:

  1. crop_image (self,img,bboxes) return three result ,but main function receive 4,crop_box have no result.

2.widerface.py ann is not defined, there is something wrong.

3.when input size =500 the res3's output is 63 ,however the res4 is 64 for resnet101 ,when excute score = score_res3 + score4 make error . dimensions is not equal 63not equal 64.

VladVin commented 5 years ago

Just upload the weights 😀

varunagrawal commented 5 years ago

@feixiangdekaka thanks for reporting the bugs. I have fixed all of them in today's refactor. Also, the README should now have everything you need to run this. If it works for you, please let me know, so I can close this issue.

@VladVin yup, need to retrain the models and make sure the accuracy is the same as the paper.

feixiangdekaka commented 5 years ago

@feixiangdekaka thanks for reporting the bugs. I have fixed all of them in today's refactor. Also, the README should now have everything you need to run this. If it works for you, please let me know, so I can close this issue.

@VladVin yup, need to retrain the models and make sure the accuracy is the same as the paper.

WIP ??? waiting in precess?haha

varunagrawal commented 5 years ago

Um Work In Progress. I have to retrain the models 😅

arki0207 commented 5 years ago

I have a bug. wider_face.py can't handle this situation： a train image has no bbx . image

varunagrawal commented 5 years ago

@arki0207 can you please try with the latest code? It should be able to handle it.

varunagrawal commented 4 years ago

@arki0207 that line makes no sense. If the number of bounding boxes is 0, why should the dataset have a line 0 0 0 0 0 0 0 0 0 0? It should be blank.

I can't understand why you would encounter this issue in the first place.

varunagrawal commented 4 years ago

Closing this issue since the original issue has been fixed. Please open new issues for new problems.