Practical overview
When it fits
Use it when you can prepare matching image and VOC-format annotation files and want a code-based workflow for training, inference, and validation.
Who it is for
- Python developers building object detectors for labeled image datasets.
- Teams that need to run predictions on images or video and assess detector performance with mAP.
Documented uses
- Train a detector for selected object labels using VOC-format annotations.
- Generate bounding-box predictions for an image or video.
- Evaluate a configured model against a validation dataset using mAP.
Requirements & limits
The repository is public and specifies the MIT license.
- The README requires pretrained backend weights in the repository root and says the code does not work without them.
- GPU use requires CUDA drivers; multi-GPU training is listed as implemented.
- The README reports testing with Python 2.7.13 and 3.5.3.
- The repository lists COCO evaluation and several alternative model backends as incomplete.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Prepare data and configuration for YOLO3 training
Set up the documented dataset structure and model configuration before training.
Before you start
- The repository and its Python dependencies, installed with
pip install -r requirements.txt. - Training and validation images with corresponding VOC-format annotations and matching filenames.
- Pretrained backend weights placed in the repository root; the README says the code does not work without these weights.
Steps
Organize images and annotations into separate training and validation folders.
Edit the JSON configuration with the folder paths, class labels, anchors, and training settings.
Optionally generate dataset anchors with the command below and copy the printed values into the configuration.
python gen_anchors.py -c config.json
Expected outcome
A configured dataset and model setup ready for the repository's training stage.
Guide limitations
- The guide covers preparation only, not a completed training run.
- The README's example dataset and configuration use raccoon detection; custom data must follow the stated folder and annotation conventions.
Run detection and evaluate a configured model
Use the repository's documented scripts to generate predictions and calculate validation-set mAP.
Before you start
- A configured repository and model weights.
- For evaluation, validation image and annotation folder paths set in the configuration.
Steps
Run the command below to process an image or video.
python predict.py -c config.json -i /path/to/image/or/videoRun the command below to compute mAP on the configured validation dataset.
python evaluate.py -c config.json
Expected outcome
The prediction script writes an image with detected bounding boxes to the same folder; the evaluation script computes mAP for the model named in the configuration.
Guide limitations
- The documented evaluation is tied to the configured validation dataset.
- The README lists COCO evaluation as incomplete.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 16 October 2026
- Original website title
- experiencor/keras-yolo3 · Official homepage title · retained in its source language