Practical overview
When it fits
Use Darkflow when a project needs a documented path from Darknet model files to TensorFlow inference or training, especially for image-folder detection, custom-label workflows, or preparing a graph for mobile integration.
Who it is for
- Developers integrating YOLO or other Darknet models into TensorFlow workflows
- Computer-vision practitioners running object detection on image folders
- Teams adapting pretrained detection models to custom datasets
- Mobile developers preparing a TensorFlow graph for app integration
Documented uses
- Convert Darknet model definitions and trained weights for use with TensorFlow.
- Run image predictions and save detections with labels, confidence values and bounding-box coordinates.
- Train or fine-tune a model using custom dataset and annotation paths.
- Export a constant TensorFlow graph definition for mobile integration.
Requirements & limits
The repository lists a GPL-3.0 license. Its README documents installation through in-place Cython extension building or pip.
- The listed dependencies are Python 3, TensorFlow 1.0, NumPy and OpenCV 3.
- The image inference workflow requires a model configuration and trained weights.
- Custom training uses dataset and annotation paths; training outputs are saved as TensorFlow checkpoints.
- The README describes mobile graph export but does not specify mobile runtimes or a complete deployment procedure.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Run image predictions with a YOLO model
Process a folder of images with a model configuration and weights, with predictions saved as JSON.
Before you start
- Python 3, TensorFlow 1.0, NumPy and OpenCV 3
- A Darkflow installation
- A model configuration and matching trained weights
- An image folder
Steps
Install Darkflow using one of the documented methods: build Cython extensions in place, use pip editable mode, or install with pip.
Run
flow --imgdir sample_img/ --model cfg/tiny-yolo.cfg --load bin/tiny-yolo.weights --json, substituting the image folder, configuration and weights you intend to use.
Expected outcome
Darkflow writes prediction output to the image folder’s `out` directory by default; JSON predictions contain labels, confidence values and bounding-box coordinates.
Guide limitations
- The example uses paths from the repository and assumes the corresponding model configuration and weights are available.
Fine-tune a model with a custom dataset
Start training a YOLO configuration from pretrained weights using dataset and annotation paths.
Before you start
- Python 3, TensorFlow 1.0, NumPy and OpenCV 3
- A Darkflow installation
- A model configuration, pretrained weights, training dataset and annotations
Steps
Set the model configuration and pretrained weights with
--modeland--load.Specify the training dataset and annotation paths with
--datasetand--annotation.Add
--trainto start training; the repository also shows resuming from the most recent checkpoint with--load -1.
Expected outcome
Darkflow trains the configured model and periodically saves intermediate TensorFlow checkpoints in `ckpt/`.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 15 October 2026
- Original website title
- thtrieu/darkflow · Official homepage title · retained in its source language