How hoangvu123456/techjam_2025 works
Project type and purpose
• This is a Python-based machine-learning pipeline for object detection using PyTorch and Faster R-CNN variants. The repository provides modules for data preparation, model construction, training, evaluation, and inference (including ensemble and video inference).
File structure
• Top level
• train.py, inference.py, inference_video.py, ensemble_inference.py, parallel.py — CLI entry points for train / single-image inference / video inference / ensemble inference / parallel execution.
• datasets.py — custom Dataset definitions.
• requirements.txt — dependencies (PyTorch, torchvision, etc.).
• data & data_configs
• data/split.py — scripts for splitting raw data.
• data_configs/ppe.yaml — YAML configs for different datasets or experiments.
• models
• create_fasterrcnn_model.py — factory that wires up various backbones.
• fasterrcnn_resnet*.py — per-backbone model definitions.
• torch_utils
• engine.py, utils.py, coco_utils.py, coco_eval.py — training/evaluation engine, utility functions, COCO API wrappers.
• utils
• transforms.py, annotations.py, general.py, logging.py — helper functions for data transforms, annotation parsing, logging, general utilities.
README(s) for additional insights
• The main README.md is empty, but torch_utils/README.md likely describes the training loop API (prepare data loader, training step, evaluation step).