How abhishekudeniyan/steel-defect-detection-on-optimize-yolo works
This repository is a steel-defect detection research project built around custom YOLO/PyTorch layers and repeatable experiments. The main user workflow is to prepare or select a dataset and model configuration, run seeded training and validation, then collect detection and performance metrics. Separate scripts support dataset inspection and model-architecture checks. The README provides only the project title, and several data-preparation scripts were read without excerpts; their exact runtime connections are therefore not asserted.
Main entry points: run_exp.py Experiment runner · register_modules.py Module registration · __init__.py Package API · custom_modules.py Custom blocks
Components 20
Dataset workflow
- Dataset preparation
scripts/prepare_dataset.py - GC10 preparation
scripts/prepare_gc10_dataset.py - V2 preparation
scripts/V2_prepare_dataset_v2.py - Dataset verification
scripts/verify_dataset.py - Dataset analysis
scripts/dataset_info.py - Sample visualization
scripts/visualize_samples.py - GC10 dataset
data/GC10_DET_YOLO/data.yaml
YOLO extensions
- Ultralytics YOLO
- SPD layers
src/spd_conv.py - Custom blocks
src/custom_modules.py - Module registration
src/register_modules.py - Package API
src/__init__.py - Architecture verification
scripts/verify_architecture.py - Model inspection
scripts/verify_model.py
Experiments and metrics
- Experiment config
configs/base.yaml - Data config
configs/data.yaml - Experiment runner
scripts/run_exp.py - Metric logger
src/logger.py - Metric CSVs
results/metrics
Other components
- Researcher
Connections 24
- Researcher → Experiment runner: runs
- Experiment runner → Experiment config: reads
scripts/run_exp.py - Experiment runner → Module registration: registers
scripts/run_exp.py - Experiment runner → Ultralytics YOLO: trains and validates
scripts/run_exp.py - Experiment runner → Metric logger: logs metrics
scripts/run_exp.py - Metric logger → Metric CSVs: writes CSV
src/logger.py - Module registration → SPD layers: registers
src/register_modules.py - Module registration → Custom blocks: registers
src/register_modules.py - Package API → SPD layers: imports
src/__init__.py - Package API → Custom blocks: imports
src/__init__.py - Package API → Metric logger: imports
src/__init__.py - Package API → Module registration: registers
src/__init__.py - Custom blocks → SPD layers: uses SPD2
src/custom_modules.py - Researcher → Dataset preparation: prepares data
- Researcher → GC10 preparation: prepares data
- Researcher → V2 preparation: prepares data
- Researcher → Dataset analysis: inspects data
- Researcher → Dataset verification: checks data
- Researcher → Sample visualization: views samples
- Researcher → Architecture verification: checks architecture
scripts/verify_architecture.py - Architecture verification → Module registration: registers modules
scripts/verify_architecture.py - Researcher → Model inspection: inspects model
- Model inspection → SPD layers: imports layers
scripts/verify_model.py - Model inspection → Custom blocks: imports blocks
scripts/verify_model.py