Original YOLO, translated to TensorFlow

What it does

This repo translates the original YOLO object detector into TensorFlow, covering both training and inference. It expects the Pascal VOC dataset and a specific YOLO_small weight file to initialize the network. Model behavior is controlled through yolo/config.py.

The interesting bit

This is a straight port, not a redesign. Its main utility is providing the original YOLO training and test pipeline entirely in TensorFlow, which makes it a reference implementation if you need the 2015 architecture reproduced faithfully in one framework.

Key highlights

• Implements the original YOLO architecture from the linked paper in TensorFlow.
• Provides both training (train.py) and testing (test.py) scripts.
• Depends only on TensorFlow and OpenCV.
• Requires downloading the YOLO_small weight file and placing it in data/weight.

Caveats

• The README offers no accuracy metrics, speed benchmarks, or guidance on expected performance.
• Configuration requires hand-editing yolo/config.py and manually obtaining weights from an external Google Drive link.
• TensorFlow version compatibility is unspecified.

Verdict

Useful if you specifically need the original YOLO spec in TensorFlow for coursework or legacy model archaeology. Everyone else should probably reach for a modern, maintained detector.