kenshohara / 3D-ResNets-PyTorch

3D ResNets for Action Recognition (CVPR 2018)
MIT License
3.9k stars 932 forks source link

What must be dimensions of frame #220

Open ArjunPukale opened 4 years ago

ArjunPukale commented 4 years ago

What must be the size of each frame of the video? Or what must be the dimensions of a batch input?

guilhermesurek commented 4 years ago

Hello, The default sample size is 112 x 112 (check opts.py flag --sample_size). The size of the batch input depends on the ram capacity, if using gpu, depends on your gpu ram capacity. The temporal window is fixed on 16 frames, so every batch will have this format: batch_size x 16 (frames) x 3 channels (RGB) x 112 x 112.