search

facebookresearch / VMZ

VMZ: Model Zoo for Video Modeling
Apache License 2.0
1.04k stars 157 forks source link

Finetune using OpticalFlow Models #7

Closed murilovarges closed 6 years ago

murilovarges commented 6 years ago

Hi @dutran,

First, thank you for making public this wonderful work.

I've already been able to use this code to finetune R(2+1)D model on my dataset and run dense prediction with success.

Now I would like to use Optical Flow for computing accuracy rate of R(2+1)D-Two-Stream and I have some questions: 1 - Can you provide the R(2+1)D-Flow trained in "Kinetics", because I found only trained "Sports1M+Kinetics"? 2 - Can you provide an example or tutorial that give more details of how to finetune Optical flow model, I notice that Optical flow models need a lot of parameters to train: frame_gap_of, sampling_rate_of, clip_length_of, flow_data_type, do_flow_aggregation, flow_alg_type?

Best regards, Murilo Varges

dutran commented 6 years ago

@murilovarges

  1. https://www.dropbox.com/s/qd0wjvhtqt88qx4/r2.5d_d34_l32_from_scratch_of_bn0.9.pkl?dl=0

  2. For the parameters of optical flows, you can check them here https://github.com/facebookresearch/R2Plus1D/blob/master/scripts/extract_feature_kinetics_of.sh

swap2ag commented 5 years ago

@murilovarges

  1. https://www.dropbox.com/s/qd0wjvhtqt88qx4/r2.5d_d34_l32_from_scratch_of_bn0.9.pkl?dl=0
  2. For the parameters of optical flows, you can check them here https://github.com/facebookresearch/R2Plus1D/blob/master/scripts/extract_feature_kinetics_of.sh

What does the file in 1 contains? Does it contain the weights of R(2+1)D model trained from scratch using kinetics flows?