zhouzhi

you can use VGGFace or VGGFace2 datasets to train the model from scratch, just like multi classifications.

I guess there is no triplet loss used in training the model. Just use softmax loss for classification and then we can remove top layers to use features for generic...

> On linux/ubuntu you need to have python dev installed, for example sudo apt install python3.10-dev, replace with your python version mark

No practise, but I think it better to train 2 models for pedstrain and vehicles separately, because these 2 classes is unsimilar at all. You can train only 1 model...

https://github.com/nwojke/cosine_metric_learning/blob/master/train_app.py#L651 maybe you can set monitor_xx as False when using cosine-softmax loss mode. the code can also monitor other loss value when using cosine-softmax mode.

There is simple way to train the feature-extraction model using VeRi, just treat it as multi classes task. Say using resnet50 in keras and add a Dense layer with softmax...

Someone has trained the mode on VeRi dataset besides mars and market1501 https://github.com/nwojke/cosine_metric_learning/issues/31 You need prepare your dataset refer to the datasets above.

参考： https://github.com/sherlockchou86/VideoPipe/issues/40#issuecomment-2069606083

modify 2 codes: https://github.com/k0suke-murakami/object_tracking/blob/master/src/box_fitting.cpp#L175 check if x/y out range of width/height https://github.com/k0suke-murakami/object_tracking/blob/master/src/imm_ukf_jpda.cpp#L97 comment this line

hello, did you implement this idea? i also need this solution. thanks!