How flame-chasers/gtr works
GTR+ trains unsupervised text-based person-search models from generated pseudo descriptions, then evaluates image–text retrieval on person-search benchmarks. The README documents three-tier description generation, confidence-weighted learning, and LargeFine-Person pretraining; the graph preserves these stages and the BLIP/vision-language internals. Some dataset and filtering implementations were not excerpted, so their detailed interactions are omitted; documented workflow edges use the README as evidence.
Main entry points: train_ps.py Confidence-weighted training · __init__.py Dataset and loader factory · blip.py BLIP vision-language · ps_filter.py Description generation
Components 15
Data preparation
- LargeFine pretraining data
data/ps_pretrain_dataset.py - Person-search datasets
data/ps_dataset.py - Dataset and loader factory
data/__init__.py
Pseudo supervision
- Description generation
ps_filter.py - Filter datasets
data/ps_filter_dataset.py
Model learning
- Pseudo-supervised pretraining
pretrain_ps.py - Confidence-weighted training
train_ps.py - Retrieval evaluation
train_ps.py - GMM retrieval model
models/blip_ps_gmm.py - Person-search model
models/blip_ps.py
Model internals
- BLIP vision-language
models/blip.py - Vision transformer
models/vit.py - Text encoder
models/med.py
Other components
- Researcher
- Ranked person matches
Connections 18
- Researcher → Description generation: generates descriptions
README.md - Description generation → Filter datasets: uses filtered data
- Description generation → LargeFine pretraining data: creates pseudo supervision
README.md - Researcher → Pseudo-supervised pretraining: starts pretraining
README.md - Researcher → Confidence-weighted training: starts training
README.md - Pseudo-supervised pretraining → Dataset and loader factory: loads datasets
pretrain_ps.py - Confidence-weighted training → Dataset and loader factory: loads datasets
train_ps.py - Dataset and loader factory → LargeFine pretraining data: selects dataset
data/__init__.py - Dataset and loader factory → Person-search datasets: selects dataset
data/__init__.py - Pseudo-supervised pretraining → Person-search model: trains model
pretrain_ps.py - Confidence-weighted training → GMM retrieval model: trains model
train_ps.py - Retrieval evaluation → GMM retrieval model: evaluates model
train_ps.py - Confidence-weighted training → Retrieval evaluation: runs evaluation
README.md - Retrieval evaluation → Ranked person matches: produces rankings
README.md - GMM retrieval model → BLIP vision-language: builds on BLIP
models/blip_ps_gmm.py - Person-search model → BLIP vision-language: builds on BLIP
models/blip_ps.py - BLIP vision-language → Vision transformer: creates vision encoder
models/blip.py - BLIP vision-language → Text encoder: creates text encoder
models/blip.py