How siyuvision/ia4u-deep-dive works
This repository packages three reproducible computer-vision experiments: whole-slide pathology detection comparison, multimodal renal-tubule box evaluation, and nnFoundation CT tumor segmentation. The user-facing pattern is to run an experiment, obtain model outputs, evaluate them, and export or report results. The CT pilot’s stage order is explicit in its pipeline runner; several other subsystem connections are documented only at the README or directory level. Edges involving unsampled behavior are omitted or marked unverified rather than inferred from filenames.
Main entry points: run_pipeline.py Pilot orchestrator · fourshot_config.py Four-shot protocol · evaluate_runs.py Run evaluation · project_env.py Project environment
Components 31
Pathology detection
- WSI inference
projects/kongnet-vs-classpose/scripts/run_kongnet_wsi_windows.py - ClassPose inference
projects/kongnet-vs-classpose/scripts/run_classpose_wsi.py - Patch inference
projects/kongnet-vs-classpose/scripts/run_patch.py - WSI preparation
projects/kongnet-vs-classpose/scripts/prepare_wsi.py - Point evaluation
projects/kongnet-vs-classpose/scripts/evaluate_points.py - QuPath export
projects/kongnet-vs-classpose/scripts/export_qupath.py
Renal box benchmark
- Model calls
projects/mllm-renal-tubule-bbox/repo/scripts/run_models.py - API client
projects/mllm-renal-tubule-bbox/repo/src/mllm_bbox/openrouter.py - Call records
projects/mllm-renal-tubule-bbox/repo/scripts/run_models.py - Run evaluation
projects/mllm-renal-tubule-bbox/repo/scripts/evaluate_runs.py - Response parsing
projects/mllm-renal-tubule-bbox/repo/src/mllm_bbox/parsing.py - Box geometry
projects/mllm-renal-tubule-bbox/repo/src/mllm_bbox/boxes.py - Box scoring
projects/mllm-renal-tubule-bbox/repo/src/mllm_bbox/evaluation.py - Cost and speed
projects/mllm-renal-tubule-bbox/repo/src/mllm_bbox/efficiency.py - Result rendering
projects/mllm-renal-tubule-bbox/repo/src/mllm_bbox/render.py - QuPath export
projects/mllm-renal-tubule-bbox/repo/scripts/export_qupath.py
CT segmentation
- Pilot orchestrator
projects/nnfoundation-tumseg-ct/repo/run_pipeline.py - Data planning
projects/nnfoundation-tumseg-ct/repo/plan_data.py - Model training
projects/nnfoundation-tumseg-ct/repo/train.py - CT prediction
projects/nnfoundation-tumseg-ct/repo/predict.py - Segmentation evaluation
projects/nnfoundation-tumseg-ct/repo/evaluate.py - Four-shot protocol
projects/nnfoundation-tumseg-ct/repo/fourshot_config.py - Four-shot preparation
projects/nnfoundation-tumseg-ct/repo/prepare_fourshot.py - Four-shot run
projects/nnfoundation-tumseg-ct/repo/run_fourshot.py - Four-shot evaluation
projects/nnfoundation-tumseg-ct/repo/evaluate_fourshot.py - Spatial prior guard
projects/nnfoundation-tumseg-ct/repo/spatial_guard.py - Scientific assets
projects/nnfoundation-tumseg-ct/repo/render_assets.py - Four-shot report
projects/nnfoundation-tumseg-ct/repo/report_fourshot.py - Project environment
projects/nnfoundation-tumseg-ct/repo/project_env.py
Other components
- Researcher
- OpenRouter API
Connections 38
- Researcher → WSI inference: runs inference
- Researcher → ClassPose inference: runs inference
- Researcher → Model calls: runs benchmark
- Researcher → Pilot orchestrator: runs experiment
- WSI inference → QuPath export: exports points
projects/kongnet-vs-classpose/scripts/run_kongnet_wsi_windows.py - ClassPose inference → QuPath export: exports points
projects/kongnet-vs-classpose/scripts/run_classpose_wsi.py - Patch inference → QuPath export: exports points
projects/kongnet-vs-classpose/scripts/run_patch.py - Model calls → API client: calls client
projects/mllm-renal-tubule-bbox/repo/scripts/run_models.py - API client → OpenRouter API: sends requests
projects/mllm-renal-tubule-bbox/repo/src/mllm_bbox/openrouter.py - Model calls → Call records: records calls
projects/mllm-renal-tubule-bbox/repo/scripts/run_models.py - Run evaluation → Box geometry: uses geometry
projects/mllm-renal-tubule-bbox/repo/scripts/evaluate_runs.py - Run evaluation → Box scoring: scores answers
projects/mllm-renal-tubule-bbox/repo/scripts/evaluate_runs.py - Run evaluation → Cost and speed: summarizes efficiency
projects/mllm-renal-tubule-bbox/repo/scripts/evaluate_runs.py - Run evaluation → Result rendering: renders results
projects/mllm-renal-tubule-bbox/repo/scripts/evaluate_runs.py - Box scoring → Response parsing: parses text
projects/mllm-renal-tubule-bbox/repo/src/mllm_bbox/evaluation.py - Box scoring → Box geometry: converts and scores
projects/mllm-renal-tubule-bbox/repo/src/mllm_bbox/evaluation.py - Cost and speed → API client: reads latency
projects/mllm-renal-tubule-bbox/repo/src/mllm_bbox/efficiency.py - Result rendering → Box geometry: uses boxes
projects/mllm-renal-tubule-bbox/repo/src/mllm_bbox/render.py - Researcher → QuPath export: exports results
- Pilot orchestrator → Data planning: runs stage
projects/nnfoundation-tumseg-ct/repo/run_pipeline.py - Pilot orchestrator → Model training: runs stage
projects/nnfoundation-tumseg-ct/repo/run_pipeline.py - Pilot orchestrator → CT prediction: runs stage
projects/nnfoundation-tumseg-ct/repo/run_pipeline.py - Pilot orchestrator → Segmentation evaluation: runs stage
projects/nnfoundation-tumseg-ct/repo/run_pipeline.py - Four-shot protocol → Project environment: uses settings
projects/nnfoundation-tumseg-ct/repo/fourshot_config.py - Four-shot preparation → Four-shot protocol: uses protocol
projects/nnfoundation-tumseg-ct/repo/prepare_fourshot.py - Four-shot preparation → Spatial prior guard: uses prior guard
projects/nnfoundation-tumseg-ct/repo/prepare_fourshot.py - Four-shot run → Four-shot protocol: uses protocol
projects/nnfoundation-tumseg-ct/repo/run_fourshot.py - Four-shot evaluation → Four-shot protocol: uses protocol
projects/nnfoundation-tumseg-ct/repo/evaluate_fourshot.py - Four-shot evaluation → Segmentation evaluation: reuses evaluation
projects/nnfoundation-tumseg-ct/repo/evaluate_fourshot.py - Four-shot evaluation → Spatial prior guard: uses prior guard
projects/nnfoundation-tumseg-ct/repo/evaluate_fourshot.py - Four-shot report → Four-shot protocol: uses protocol
projects/nnfoundation-tumseg-ct/repo/report_fourshot.py - Four-shot report → Scientific assets: uses assets
projects/nnfoundation-tumseg-ct/repo/report_fourshot.py - Researcher → Four-shot preparation: prepares cohort
- Researcher → Four-shot run: runs retest
- Researcher → Four-shot report: reviews evidence
- Model training → Project environment: configures runtime
projects/nnfoundation-tumseg-ct/repo/train.py - CT prediction → Project environment: configures runtime
projects/nnfoundation-tumseg-ct/repo/predict.py - Segmentation evaluation → Project environment: uses paths
projects/nnfoundation-tumseg-ct/repo/evaluate.py