Kumara Kahatapitiya

In my setting, accimage was significantly faster than Pillow (~1.4x-1.5x), but not so much compared to Pillow-SIMD. Roughly speaking, for every 1000 images Pillow-SIMD loads and pre-processes (crop+resize+transpose), accimage does...

During training split_bn parameters (eg: self.split_bn.running_mean.data) inside SubBatchNorm will be updated, and they will be copied to bn parameters (eg: self.bn.running_mean.data) during eval, by running https://github.com/kkahatapitiya/X3D-Multigrid/blob/d63d8fe6210d2b38aa26d71b0062b569687d6be2/train_x3d_kinetics_multigrid.py#L205 Are you doing this?...

You have to run aggregate_sub_bn_stats(), before validation (i.e., when you put the model in eval() mode) everytime.

Hi, Thank you for your interest in our work and sorry about the confusion. It is true that we use T=64 in the Coarse stream and T=128 in the Fine...

Sorry about the delay in response. The purpose of using `testing` flag for both train/val splits when extracting features, is to avoid any random sampling and augmentations that apply when...

Yes, Kinetics experiments are trained from scratch. Input clips are provided as RGB frames. You can refer to the dataset file here: https://github.com/kkahatapitiya/X3D-Multigrid/blob/d63d8fe6210d2b38aa26d71b0062b569687d6be2/kinetics.py#L161

Yes, as long as the the dataset is large-enough, you can train form scratch on your data. Otherwise, I would suggest to finetune the K400 pretrained model on your data....

Sorry about the long delay in response. I compute top-1 accuracy (video-level) here: https://github.com/kkahatapitiya/X3D-Multigrid/blob/d63d8fe6210d2b38aa26d71b0062b569687d6be2/train_x3d_kinetics_multigrid.py#L263 In multi-crop testing, multiple crops are reshaped into batch dimension as in: https://github.com/kkahatapitiya/X3D-Multigrid/blob/d63d8fe6210d2b38aa26d71b0062b569687d6be2/train_x3d_kinetics_multigrid.py#L242 It can handle...

Sorry about the long delay in response. Since the data split and multiple training hyperparameters are different, I am not sure what the expected performance would look like. If you...

Sorry about the delay in response. X3D uses split batchnorm to compute batchnorm stats with a constant batch size, no matter the actual input batch size. To do so, the...