search

yxgeee / FD-GAN

[NeurIPS-2018] FD-GAN: Pose-guided Feature Distilling GAN for Robust Person Re-identification.
https://yxgeee.github.io/projects/fdgan.html
281 stars 80 forks source link

How to use this model in my own videos? #15

Open cppntn opened 5 years ago

cppntn commented 5 years ago

Hello, thanks for this fantastic work. How can I use this model to recognize people in my own video? I have a detection network that detects bounding box of people, what would you recommend in order to use fd-gan to assign unique ID's to people in the video?

Thanks

yxgeee commented 5 years ago

Maybe you can use the pretrained model to extract the visual features for each person image, and then cluster them into several groups by calculating the distances between each other.

cppntn commented 5 years ago

Thanks @yxgeee. How can I use a pretrained model just for the inference? What is the model and the weights that I should use?

MOreover, what kind of clustering algorithms would you recommend for calculating the distances between features?

cppntn commented 5 years ago

Hello @yxgeee, sorry to bother you.. How can I extract the feature from an image of a person? I mean, once I have loaded the weights on the SiameseNet class (should the weights be "best_net_E.pth"?) how can I do inference? have I to forward only the base model (resnet50) or the embed model too? Can you tell me a little more about this? A great thanks to you