bbvisual/countex

How bbvisual/countex works

CountEx is a fine-grained visual counting system: users specify a target and a distractor, optionally with visual exemplars, and the model predicts the target count while suppressing the distractor. Its main workflow is data preparation, model inference and training, then evaluation. The repository also contains a separate count-only-supervision extension, with density-only and detection-prior/alternating-training variants. Edges reflect the README and shown source references; unsampled subsystem details and links are omitted where their wiring is not evidenced.

Main entry points: CountEX.py CountEx Model · train.py Training Entry · utils.py Data Preparation · CountEXStage2.py Detection-Prior Model

Components 21

CountEx Core

Data Preparation src/utils.py
CountEx Model src/hf_model/CountEX.py
Matching Losses src/criterion.py
Fine-Grained Trainer src/trainer.py
Training Entry src/train.py

Count-Only Extension

Count-Only Data Tools count_only/src
Count-Only Model Variants count_only/src/hf_model
Density-Only Model count_only/src/hf_model/CountEXDensityOnly.py
Detection-Prior Model count_only/src/hf_model/CountEXStage2.py
Alternating Trainer count_only/src/alternating_trainer.py
Count-Only Training count_only/src/train.py
Alternating Training Entry count_only/src/train_alternating.py

Modeling Infrastructure

Grounding DINO src/hf_model/modeling_grounding_dino.py

Evaluation

Core Evaluation src/eval.py
Count-Only Evaluation count_only/src/eval.py
Density Evaluation count_only/src/eval_density.py
Localization Evaluation count_only/src/eval_localization.py
Matched Evaluations count_only/src/eval_matched.py

Other components

Researcher
CoCount Dataset
PairTally Benchmark

Connections 29

Researcher → CoCount Dataset: selects data README.md
CoCount Dataset → Training Entry: trains on README.md
Researcher → CountEx Model: specifies query README.md
Data Preparation → Fine-Grained Trainer: prepares batches src/trainer.py
Training Entry → CountEx Model: loads model src/train.py
Training Entry → Matching Losses: configures loss src/train.py
Training Entry → Fine-Grained Trainer: creates trainer src/train.py
Training Entry → Data Preparation: uses utilities src/train.py
CountEx Model → Grounding DINO: extends backbone src/hf_model/CountEX.py
Fine-Grained Trainer → Data Preparation: uses utilities src/trainer.py
Core Evaluation → CountEx Model: evaluates model src/eval.py
Core Evaluation → Data Preparation: uses utilities src/eval.py
Researcher → PairTally Benchmark: selects benchmark README.md
PairTally Benchmark → Core Evaluation: evaluates transfer README.md
Researcher → Count-Only Data Tools: prepares supervision count_only/src/build_acc_only_dataset.py
Count-Only Training → Count-Only Model Variants: selects variants count_only/src/train.py
Count-Only Training → Fine-Grained Trainer: uses trainer count_only/src/train.py
Count-Only Training → Matching Losses: uses losses count_only/src/train.py
Count-Only Model Variants → Grounding DINO: uses backbone count_only/src/hf_model/CountEX.py
Density-Only Model → Grounding DINO: uses backbone count_only/src/hf_model/CountEXDensityOnly.py
Detection-Prior Model → CountEx Model: extends model count_only/src/hf_model/CountEXStage2.py
Alternating Training Entry → Detection-Prior Model: loads variant count_only/src/train_alternating.py
Alternating Training Entry → Alternating Trainer: creates trainer count_only/src/train_alternating.py
Alternating Trainer → Data Preparation: uses utilities count_only/src/alternating_trainer.py
Count-Only Evaluation → CountEx Model: evaluates model count_only/src/eval.py
Density Evaluation → Density-Only Model: evaluates model count_only/src/eval_density.py
Density Evaluation → Detection-Prior Model: evaluates variant count_only/src/eval_density.py
Localization Evaluation → Detection-Prior Model: evaluates variant count_only/src/eval_localization.py
Matched Evaluations → Detection-Prior Model: evaluates variant count_only/src/eval_matched.py

More Python diagrams: mrlt8/docker-wyze-bridge · zhengpeng7/birefnet · adamerose/pandasgui · taishi-i/awesome-chatgpt-repositories · geekan/scrapy-examples

Drawn by GitDiagram on September 30, 2026 from the repository's file tree, README and source files. Read as Markdown ·