Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Update S3D weights #6537

Merged
merged 3 commits into from
Sep 5, 2022
Merged

Update S3D weights #6537

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
52575a0
S3D weight deployment
datumbox Sep 5, 2022
81a153b
Update accuracies.
datumbox Sep 5, 2022
3ca5ac4
Address review comments.
datumbox Sep 5, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
16 changes: 16 additions & 0 deletions references/video_classification/README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -81,6 +81,7 @@ Video resnet models:
```
# number of frames per clip
--clip_len 16 \
--frame-rate 15 \
# allow for temporal jittering
--clips_per_video 5 \
--batch-size 24 \
Expand All @@ -97,6 +98,21 @@ Video resnet models:
--val-crop-size 112 112
```

### S3D

The S3D model was trained similarly to the above but with the following changes on the default configuration:
datumbox marked this conversation as resolved.
Show resolved Hide resolved
```
--batch-size=12 --lr 0.2 --clip-len 64 --clips-per-video 5 --sync-bn \
--train-resize-size 256 256 --train-crop-size 224 224 --val-resize-size 256 256 --val-crop-size 224 224
```

We used 64 GPUs to train the architecture.

To estimate the validation statistics of the model, we run the reference script with the following configuration:
```
--batch-size=16 --test-only --clip-len 128 --clips-per-video 1
```

### Additional video modelling resources

- [Video Model Zoo](https://github.com/facebookresearch/VMZ)
Expand Down
14 changes: 6 additions & 8 deletions torchvision/models/video/s3d.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -104,7 +104,7 @@ class S3D(nn.Module):
def __init__(
self,
num_classes: int = 400,
dropout: float = 0.0,
dropout: float = 0.2,
norm_layer: Optional[Callable[..., torch.nn.Module]] = None,
) -> None:
super().__init__()
Expand Down Expand Up @@ -153,28 +153,26 @@ def forward(self, x):

class S3D_Weights(WeightsEnum):
KINETICS400_V1 = Weights(
url="https://download.pytorch.org/models/s3d-1bd8ae63.pth",
url="https://download.pytorch.org/models/s3d-d76dad2f.pth",
transforms=partial(
VideoClassification,
crop_size=(224, 224),
resize_size=(256, 256),
mean=(0.5, 0.5, 0.5),
std=(0.5, 0.5, 0.5),
),
meta={
"min_size": (224, 224),
"min_temporal_size": 14,
"categories": _KINETICS400_CATEGORIES,
"recipe": "https://github.com/pytorch/vision/pull/6412#issuecomment-1219687434",
"recipe": "https://github.com/pytorch/vision/tree/main/references/video_classification#s3d",
"_docs": (
"The weights are ported from a community repository. The accuracies are estimated on clip-level "
"The weights aim to approximate the accuracy of the paper. The accuracies are estimated on clip-level "
"with parameters `frame_rate=15`, `clips_per_video=1`, and `clip_len=128`."
),
"num_params": 8320048,
"_metrics": {
"Kinetics-400": {
"acc@1": 67.315,
"acc@5": 87.593,
"acc@1": 68.368,
"acc@5": 88.050,
}
},
},
Expand Down