search

openvinotoolkit / training_extensions

Train, Evaluate, Optimize, Deploy Computer Vision Models via OpenVINO™
https://openvinotoolkit.github.io/training_extensions/
Apache License 2.0
1.14k stars 443 forks source link

person-reidentification-retail #159

Closed dinara92 closed 4 years ago

dinara92 commented 5 years ago

First of all, thank you very much for open sourcing this repo.

I have a question about pytorch_toolkit/face_recognition/* code.

I would like to re-train (and/or fine-tune) a "person-reidentification-retail" Openvino model, on a custom dataset (along with Market1501 dataset). I know that it uses RMNet as a backbone, same as "face-recognition-model".

I wonder if I can use code in pytorch_toolkit/face_recognition/* for this purpose.

AlexanderDokuchaev commented 5 years ago

Hi @dinara92

Unfortunately, person-reidentification-retail-0079 model was trained in Caffe and we dont planing to provide training code and checkpoint for this model.

To train person-reid by using pytorch_toolkit/face_recognition code is possible but you should create CustomDataset class (like this) without alignment by landmarks and remove all code that use landmarks in all other places.

To select RMNet backbone you should use '--model rmnet' argument.

AlexanderDokuchaev commented 5 years ago

@dinara92 Some good news, we are planning publish code to train person-reid model in near future.

dinara92 commented 5 years ago

many thanks for a thorough reply! As I found in this issue, I assume RMNet model pre-trained on the OpenImage dataset may not be released?

AlexanderDokuchaev commented 5 years ago

This is true for now :disappointed:

whatisor commented 5 years ago

@dinara92 Some good news, we are planning publish code to train person-reid model in near future.

When will it come true?

AlexanderDokuchaev commented 4 years ago

@dinara92 @whatisor pytorch_toolkit/person_reidentification