predict_transform=make_transform(val_post_tensor_transform),
batch_size=8,
clip_sampler="uniform",
clip_duration=1,
video_sampler=RandomSampler,
decode_audio=False,
num_workers=8)
# 4. List the available models
print(VideoClassifier.available_backbones())
# out: ['efficient_x3d_s', 'efficient_x3d_xs', ... ,slowfast_r50', 'x3d_m', 'x3d_s', 'x3d_xs']
print(VideoClassifier.get_backbone_details("x3d_xs"))
# 5. Build the VideoClassifier with a PyTorchVideo backbone.
model = VideoClassifier(backbone="x3d_xs",
num_classes=datamodule.num_classes,
serializer=Labels(),
pretrained=False)
# 6. Finetune the model
trainer = flash.Trainer(fast_dev_run=True)
trainer.finetune(model, datamodule=datamodule, strategy=NoFreeze())
trainer.save_checkpoint("video_classification.pt")
# 7. Make a prediction
predictions = model.predict(
os.path.join(flash.PROJECT_ROOT, "data/kinetics/predict"))
print(predictions)
# ['marching', 'flying_kite', 'archery', 'high_jump', 'bowling']
from flash.core.data.utils import download_data
from flash.video import VideoClassificationData, VideoClassifier
# 1. Create the DataModule
# Find more datasets at https://pytorchvideo.readthedocs.io/en/latest/data.html
download_data("https://pl-flash-data.s3.amazonaws.com/kinetics.zip", "./data")
datamodule = VideoClassificationData.from_folders(
train_folder=os.path.join(os.getcwd(), "data/kinetics/train"),
val_folder=os.path.join(os.getcwd(), "data/kinetics/val"),
clip_sampler="uniform",
clip_duration=1,
decode_audio=False,
)
# 2. Build the task
model = VideoClassifier(backbone="x3d_xs",
num_classes=datamodule.num_classes,
pretrained=False)
# 3. Create the trainer and finetune the model
trainer = flash.Trainer(max_epochs=3, gpus=torch.cuda.device_count())
trainer.finetune(model, datamodule=datamodule, strategy="freeze")
# 4. Make a prediction
predictions = model.predict(os.path.join(os.getcwd(), "data/kinetics/predict"))
print(predictions)
# 5. Save the model!
trainer.save_checkpoint("video_classification.pt")
predict_folder=os.path.join(_PATH_ROOT, "data/kinetics/predict"),
train_transform=make_transform(train_post_tensor_transform),
val_transform=make_transform(val_post_tensor_transform),
predict_transform=make_transform(val_post_tensor_transform),
batch_size=8,
clip_sampler="uniform",
clip_duration=2,
video_sampler=RandomSampler,
decode_audio=False,
)
# 4. List the available models
print(VideoClassifier.available_models())
# out: ['efficient_x3d_s', 'efficient_x3d_xs', ... ,slowfast_r50', 'x3d_m', 'x3d_s', 'x3d_xs']
print(VideoClassifier.get_model_details("x3d_xs"))
# 5. Build the model - `x3d_xs` comes with `nn.Softmax` by default for their `head_activation`.
model = VideoClassifier(model="x3d_xs", num_classes=datamodule.num_classes)
model.serializer = Labels()
# 6. Finetune the model
trainer = flash.Trainer(max_epochs=3)
trainer.finetune(model, datamodule=datamodule, strategy=NoFreeze())
trainer.save_checkpoint("video_classification.pt")
# 7. Make a prediction
predictions = model.predict(
os.path.join(_PATH_ROOT, "data/kinetics/predict"))
print(predictions)
video_sampler=RandomSampler,
decode_audio=False,
train_transform=make_transform(train_post_tensor_transform),
val_transform=make_transform(val_post_tensor_transform),
predict_transform=make_transform(val_post_tensor_transform),
num_workers=8,
batch_size=8,
)
# 4. List the available models
print(VideoClassifier.available_models())
# out: ['efficient_x3d_s', 'efficient_x3d_xs', ... ,slowfast_r50', 'x3d_m', 'x3d_s', 'x3d_xs']
print(VideoClassifier.get_model_details("x3d_xs"))
# 5. Build the model - `x3d_xs` comes with `nn.Softmax` by default for their `head_activation`.
model = VideoClassifier(model="x3d_xs", num_classes=datamodule.num_classes)
model.serializer = Labels()
# 6. Finetune the model
trainer = flash.Trainer(max_epochs=3, gpus=1)
trainer.finetune(model, datamodule=datamodule, strategy=NoFreeze())
trainer.save_checkpoint("video_classification.pt")
# 7. Make a prediction
val_folder = os.path.join(
_PATH_ROOT, os.path.join(_PATH_ROOT, "data/kinetics/predict"))
predictions = model.predict(
[os.path.join(val_folder, f) for f in os.listdir(val_folder)])
print(predictions)
