How vamsi419-cell/uncertanity-adpative-kalam-fusion-for-open-vocabulary-multi-object-detection works
This repository combines the OVT-B open-vocabulary multi-object-tracking benchmark and an OVTrack baseline with a separate Diffusers-based image-generation package. The main benchmark workflow is to configure OVT-B class prompts and data paths, load video annotations, run a tracker, and evaluate tracking results. The tree also contains model training and multiple tracker implementations. Diffusion pipelines and their model, scheduler, and configuration layers are included as a distinct runtime branch; source evidence for several connections is limited to the SOURCE INDEX or README.
Main entry points: pipeline_ovtrack_image_generation.py OVTrack image pipeline · ovtrack.py OVTrack model · trackers Tracker family · schedulers Diffusion schedulers
Components 22
OVT-B benchmark
- OVT-B data
- OVT-B classes
data/ovtb/ovtb_classname.py - Tracking configs
configs/ovtrack-teta/ovtb/ovtrack_r50_ua.py - Video datasets
ovtrack/datasets/coco_video_dataset.py - Video annotation parser
ovtrack/datasets/parsers/coco_video_parser.py - MOT evaluation
ovtrack/core/evaluation/mot.py
OVTrack runtime
- Inference entry
tools/inference.py - Inference API
ovtrack/apis/inference.py - Model builder
ovtrack/models/builder.py - OVTrack model
ovtrack/models/mot/ovtrack.py - Detection and tracking heads
ovtrack/models/roi_heads - Tracker family
ovtrack/models/trackers - Track association
ovtrack/core/track/association.py - Adaptive motion filter
ovtrack/models/trackers/adaptive_kalman.py - Training API
ovtrack/apis/train.py
Diffusion generation
- OVTrack image pipeline
diffusers_clean/pipelines/stable_diffusion/pipeline_ovtrack_image_generation.py - Tracking-aware diffusion
diffusers_clean/pipelines/stable_diffusion/pipeline_stable_diffusion_ovtrack.py
Diffusion components
- Denoising and VAE models
diffusers_clean/models - Diffusion schedulers
diffusers_clean/schedulers - Pipeline base
diffusers_clean/pipeline_utils.py - Model configuration
diffusers_clean/configuration_utils.py
Other components
- Researcher
Connections 22
- Researcher → OVT-B data: downloads
README.md - Researcher → OVT-B classes: configures classes
README.md - Researcher → Tracking configs: sets data paths
README.md - OVT-B data → Video datasets: provides annotations
- Video datasets → Video annotation parser: uses parser
ovtrack/datasets/coco_video_dataset.py - Researcher → Inference entry: runs tracking
README.md - Inference entry → Inference API: calls inference
tools/inference.py - Inference API → Model builder: builds model
ovtrack/apis/inference.py - Model builder → OVTrack model: registers model
- OVTrack model → Detection and tracking heads: uses heads
- OVTrack model → Tracker family: dispatches tracking
- Tracker family → Track association: associates tracks
- Tracker family → Adaptive motion filter: uses motion filter
- Researcher → MOT evaluation: evaluates results
README.md - Researcher → Training API: trains baseline
- OVTrack image pipeline → Denoising and VAE models: uses models
diffusers_clean/pipelines/stable_diffusion/pipeline_ovtrack_image_generation.py - OVTrack image pipeline → Diffusion schedulers: uses scheduler
diffusers_clean/pipelines/stable_diffusion/pipeline_ovtrack_image_generation.py - OVTrack image pipeline → Pipeline base: registers components
diffusers_clean/pipelines/stable_diffusion/pipeline_ovtrack_image_generation.py - Tracking-aware diffusion → Denoising and VAE models: uses models
diffusers_clean/pipelines/stable_diffusion/pipeline_stable_diffusion_ovtrack.py - Tracking-aware diffusion → Diffusion schedulers: uses scheduler
diffusers_clean/pipelines/stable_diffusion/pipeline_stable_diffusion_ovtrack.py - Pipeline base → Model configuration: inherits configuration
diffusers_clean/pipeline_utils.py - Diffusion schedulers → Model configuration: uses configuration
diffusers_clean/schedulers/scheduling_pndm_flax.py