Practical overview
When it fits
Use it when you need a codebase to run CenterNet training and model evaluation on your own hardware, with a documented ResNet-18 COCO training example.
Who it is for
- Developers and researchers building or evaluating CenterNet object-detection models.
- Teams that can run Python and PyTorch workflows on self-hosted GPU hardware.
Documented uses
- Train CenterNet models using the repository's project configurations and supported ResNet examples.
- Test a saved or downloaded model checkpoint.
- Receive an automatically generated Markdown performance report after training.
Requirements & limits
The repository is public on GitHub and identifies its license as Apache-2.0.
- The documented stack includes Python 3.6+, PyTorch 1.3+, matching torchvision, OpenCV, pycocotools and GCC 4.9+.
- The README says at least one GPU is needed during compilation.
- ResNet-18 and ResNet-50 training is described as requiring more than 150 GB of memory; the README suggests reducing NUM_WORKER when memory is insufficient.
- The example training and checkpoint-testing commands use eight GPUs.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Train the documented ResNet-18 COCO setup
Run the README's example training command for CenterNet with a ResNet-18 backbone.
Before you start
- Python 3.6 or newer, PyTorch 1.3 or newer, matching torchvision, OpenCV and pycocotools.
- GCC 4.9 or newer and at least one GPU during compilation.
- For ResNet-18 training, the README recommends more than 150 GB of memory; it suggests reducing NUM_WORKER in config.py if memory is insufficient.
Steps
Install Cython and the COCO Python API as shown in the README.
From the repository, run the command below.
pip install -e .Change to
playground/centernet.res18.coco.512size.Run
dl_train --num-gpus 8.
Expected outcome
The README says training automatically evaluates the model and generates a README.md file reporting performance.
Guide limitations
- The example specifies eight GPUs.
- The README's high-memory guidance applies to ResNet-18 and ResNet-50 training.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 17 October 2026
- Original website title
- FateScript/CenterNet-better · Official homepage title · retained in its source language