VideoPipe
A cross-platform video structuring (video analysis) framework based on CV models & mLLM.
- 3k
- C++
- Apache-2.0
- 9 mo ago
A cross-platform video structuring (video analysis) framework based on CV models & mLLM.
High-performance multiple object tracking based on YOLO, Deep SORT, and KLT 🚀
[ECCV 2020] QAConv: Interpretable and Generalizable Person Re-Identification with Query-Adaptive Convolution and Temporal Lifting, and [CVPR 2022] GS: Graph Sampling Based Deep Metric Learning
[NeurIPS 2021] TransMatcher: Deep Image Matching Through Transformers for Generalizable Person Re-identification
Structured Output Matching & Association: online multi-person tracking built on the structured output of a multi-task wholebody detector, one whole-frame pass per frame (plus per-crop ReID embeddings in SOMA-R) — no tiling, no offline post-processing.
Temporal Lifting (TLift), a model-free temporal cooccurrence based score weighting method proposed in "Interpretable and Generalizable Person Re-Identification with Query-Adaptive Convolution and Temporal Lifting".
Person Re-identification by Local Maximal Occurrence Representation and Metric Learning