Please follow the installation instructions in INSTALL.md.
- We uniformly sample 32 frames during training and inference.
- We use 1-clip 1-crop evaluation for 3D network with the resolution of 256x256 following the original implementation.
-
TRAIN.LAMBDA_AV
denotes the coefficient$\lambda$ in the loss function and we set it to be 0.75. - We train 3D network SlowFast with 4 NVIDIA Tesla V100 (32GB) cards and the model is pretrained on Kinetics400 before training on Something-Something V1.
-
Specify the directory of datasets with
DATA.PATH_PREFIX
inexp/slowfast_sthv1/run.sh
. -
Specify the directory of output with
OUTPUT_DIR
inrun.sh
. -
Download the pretrained model on Kinetics400 from the original repo and specify the path with
CHECKPOINT_FILE_PATH
inconfigs/sth/SLOWFAST_8x8_R50_FFN.yaml
. -
Simply run the training scripts in exp as followed:
bash exp/slowfast_sthv1/run.sh ## baseline training bash exp/slowfast_sthv1/run_MCA.sh ## MCA training
-
Specify the directory of datasets with
DATA.PATH_PREFIX
inexp/slowfast_sthv1/test.sh
. -
Please download pretrained models from Google Drive.
-
Specify the directory of the pretrained model with
TEST.CHECKPOINT_FILE_PATH
intest.sh
. -
Run the inference scripts in exp as followed:
bash exp/slowfast_sthv1/test.sh