SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2311.18825 · NeurIPS · 2023

CAST: Cross-Attention in Space and Time for Video Action Recognition

Dongho Lee, Jongseo Lee, Jinwoo Choi

arXiv · PDF · Open in the Atlas

Code that ran

We lifted 17 functions out of this paper's own repositories and ran 9 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.

RepositoryRoleRan
KHU-VLL/CAST canonical 5 of 9
khu-vll/cast — 4 of 8
FunctionStatusWhere it lives
Attention Ran khu-vll/cast/models/bidir_modeling_crossattn.py
pointer only (licence: NOASSERTION) · get_code("b181227b8cc95f3d")
CrossAttentionT2S Ran khu-vll/cast/models/bidir_modeling_crossattn.py
pointer only (licence: NOASSERTION) · get_code("ef0019b3743b17db")
LayerNorm Ran khu-vll/cast/models/bidir_modeling_crossattn.py
pointer only (licence: NOASSERTION) · get_code("353426ba46a4932c")
composition_train_class_batch Ran KHU-VLL/CAST/engine_for_compomodel.py
pointer only (licence: NOASSERTION) · get_code("457807b46c3549fe")
crop_clip Ran KHU-VLL/CAST/util_tools/functional.py
pointer only (licence: NONE) · get_code("bc39b8ed5fefa7eb")
cross_train_class_batch Ran KHU-VLL/CAST/engine_for_onemodel.py
pointer only (licence: NOASSERTION) · get_code("2604141544b5eec0")
get_resize_sizes Ran KHU-VLL/CAST/util_tools/functional.py
pointer only (licence: NONE) · get_code("2b53e457d5632b4e")
get_sinusoid_encoding_table Ran khu-vll/cast/models/bidir_modeling_crossattn.py
pointer only (licence: NOASSERTION) · get_code("9b2250a3d13ed688")
resize_clip Ran KHU-VLL/CAST/util_tools/functional.py
pointer only (licence: NONE) · get_code("981936c73d3cdd12")
Adapter Not yet run khu-vll/cast/models/bidir_modeling_crossattn.py
pointer only (licence: NOASSERTION) · get_code("857af2a7de44faaa")
Block Not yet run khu-vll/cast/models/bidir_modeling_crossattn.py
pointer only (licence: NOASSERTION) · get_code("1dd28bf10048de7d")
CrossAttentionS2T Not yet run khu-vll/cast/models/bidir_modeling_crossattn.py
pointer only (licence: NOASSERTION) · get_code("445bb3da9e1748a4")
STCrossTransformer Not yet run khu-vll/cast/models/bidir_modeling_crossattn.py
pointer only (licence: NOASSERTION) · get_code("d26fe19969fbc2c2")
build_dataset Not yet run KHU-VLL/CAST/dataset/datasets.py
pointer only (licence: NOASSERTION) · get_code("bc71b534fe126be4")
clip_vit_base_patch16_224 Not yet run KHU-VLL/CAST/models/clip_modelling_finetune.py
pointer only (licence: NOASSERTION) · get_code("76274d93634ef7d9")
get_loss_scale_for_deepspeed Not yet run KHU-VLL/CAST/engine_for_compomodel.py
pointer only (licence: NONE) · get_code("29416043c7035e7c")
tensor_normalize Not yet run KHU-VLL/CAST/dataset/kinetics.py
pointer only (licence: NONE) · get_code("14b3b13b2e7ee844")

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Recognizing human actions in videos requires spatial and temporal understanding. Most existing action recognition models lack a balanced spatio-temporal understanding of videos. In this work, we propose a novel two-stream architecture, called Cross-Attention in Space and Time (CAST), that achieves a balanced spatio-temporal understanding of videos using only RGB input. Our proposed bottleneck cross-attention mechanism enables the spatial and temporal expert models to exchange information and make synergistic predictions, leading to improved performance. We validate the proposed method with extensive experiments on public benchmarks with different characteristics: EPIC-KITCHENS-100, Something-Something-V2, and Kinetics-400. Our method consistently shows favorable performance across these datasets, while the performance of existing methods fluctuates depending on the dataset characteristics. The code is available at https://github.com/KHU-VLL/CAST.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2311.18825")
get_code_for_paper("2311.18825")
have("2311.18825")

Connect an agent — have() is free.