search

mit-han-lab / temporal-shift-module

[ICCV 2019] TSM: Temporal Shift Module for Efficient Video Understanding
https://arxiv.org/abs/1811.08383
MIT License
2.05k stars 418 forks source link

about file 'mobilenetv2_jester_online.pth.tar' #187

Open tjinjin95 opened 3 years ago

tjinjin95 commented 3 years ago

Hi, thank you for providing the code, I have some questions about file 'mobilenetv2_jester_online.pth.tar'

  1. how to train 'mobilenetv2_jester_online.pth.tar'? Is its training process related to TSM?
  2. how to train a online model? (deal with buffer)
chongyangwang-song commented 3 years ago

@tjinjin95 I trained online model using the TSM code. Then I save the model and inference using the inference code. Where you need to change is the buffer size ,you need make the buffer size compatible to your input . I do this,The output seems to be right.

NB-Xie commented 3 years ago

@tjinjin95 I trained online model using the TSM code. Then I save the model and inference using the inference code. Where you need to change is the buffer size ,you need make the buffer size compatible to your input . I do this,The output seems to be right.

hello there! may i know:

did you use the Jester dataset , and did you change the bi-direction to uni-direction shift in training? appreciate!

dragen1860 commented 3 years ago

@chongyangwang-song Cools. Looks like you have implemented the training code the right way. Could you kindly share the training code for jester dataset to me: liangqu.long@gmail.com, thank you .

chongyangwang-song commented 3 years ago

我自己做的数据集，在自己数据集上跑的

---Original--- From: "Jackie @.> Date: Mon, Jul 19, 2021 11:44 AM To: @.>; Cc: @.**@.>; Subject: Re: [mit-han-lab/temporal-shift-module] about file 'mobilenetv2_jester_online.pth.tar' (#187)

@chongyangwang-song Cools. Looks like you have implemented the training code the right way. Could you kindly share the training code for jester dataset to me: @.***, thank you .

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.

sean-001 commented 3 years ago

@tjinjin95 I trained online model using the TSM code. Then I save the model and inference using the inference code. Where you need to change is the buffer size ,you need make the buffer size compatible to your input . I do this,The output seems to be right.

Hi, I also try to train the online model using the offline code, but the results are not good that the camera can't recognize hand gesture correctly. Then I try to evaluate the 'mobilenetv2_jester_online.pth.tar' parameters using the offline code, the results are not good either, the beat_prec1 is only around 5%. Thus I guess that the online code may need another special code for training.

And it seems that you have implemented the training code the right way, could you please provide some details about the training ? Or are there any problems with my policy? Thank you!