temporal-embedding-aggregation

Experimental repo for testing various ways of aggregating information across video frame embeddings

Kinetics700 Baseline results:

Zero-shot Classification:

	Accuracy
Top-1	0.31
Top-5	0.56
mean(Top1, Top5)	0.44

Linear-probe Classification:

	Accuracy
Top-1	0.41
Top-5	0.65
mean(Top1, Top5)	0.53

Notes:

EmbeddingDatasetReader isn't in the current pip install version of clip_video_encode. Please install with python setup.py install from the cloned clip_video_encode repo.

Huggingface load_dataset is not currently implemented for the CLIP-Kinetics700 dataset (iejMac/clip-video-encode#14). Please use:

git lfs install
git clone https://huggingface.co/datasets/iejMac/CLIP-Kinetics700
cd CLIP-Kinetics700
git lfs pull