HAKE: Human Activity Knowledge Engine
MVIG - Shanghai Jiao Tong University
Introduction
Human Activity Knowledge Engine (HAKE) aims at promoting activity understanding with a large-scale knowledge base and visual reasoning. To afford knowledge-based learning/reasoning, HAKE provides not only 26M+ human body part-level atomic action labels (Part States) but also holistic object knowledge annotations. It boosts the performances of several widely-used activity benchmarks (HICO, HICO-DET, V-COCO, AVA, etc), and can convert human boxes into multi-modal representations for diverse downstream tasks, e.g., image/video action recognition/detection, captioning, VQA, visual reasoning, etc. Now, the HAKE project contains the works of 8 papers (CVPR'18/19/20, NeurIPS'20, TPAMI'21).
We will keep enriching HAKE to make it a general research platform of knowledge mining, visual reasoning, and causal inference. Come and join us!
Project
1) HAKE-Image (CVPR'18/20): HAKE-HICO, HAKE-HICO-DET, HAKE-Large, Extra-40-verbs.
2) HAKE-Action-TF, HAKE-Action-Torch (CVPR'18/19/20, NeurIPS'20, TPAMI'21): SOTA action understanding methods and the corresponding HAKE-enhanced versions (TIN， IDN).
3) HAKE-3D (CVPR'20): 3D human-object representation for action understanding (DJ-RN).
4) HAKE-Object (CVPR'20, TPAMI'21): object knowledge learner to advance action understanding (SymNet).
5) HAKE-A2V (CVPR'20): Activity2Vec, a general activity feature extractor based on HAKE data, converts a human (box) to a fixed-size vector, PaSta and action scores.
6) Halpe: a joint project under Alphapose and HAKE, full-body human keypoints (body, face, hand, 136 points) of 50,000 HOI images.
7) HOI Learning List: a list of recent HOI (Human-Object Interaction) papers, code, datasets and leaderboard on widely-used benchmarks.