Practical overview
When it fits
Use the repository when you need PyTorch model code and a documented, container-based path for evaluating Scaled-YOLOv4 weights with COCO data.
Who it is for
- Computer-vision developers incorporating YOLOv4 object detection into their systems.
- Researchers and practitioners studying Scaled-YOLOv4 variants or evaluating them on COCO.
Documented uses
- Inspect implementations of YOLOv4-CSP, YOLOv4-tiny and YOLOv4-large.
- Run the documented COCO evaluation commands for P5, P6 or P7 weights.
- Compare the repository’s listed evaluation metrics across supported model sizes.
Requirements & limits
- The implementation uses Python and PyTorch and is licensed under GPL-3.0.
- The README’s setup references Docker with NVIDIA tooling, an NVIDIA PyTorch container, mish-cuda and mounted COCO data.
- The listed test commands use different image sizes for P5, P6 and P7 and specify device 0.
- The repository reports metrics for its stated evaluation setup; those figures do not establish results on different datasets or environments.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Set up the documented testing environment
Prepare the repository’s container and install mish-cuda before following its testing instructions.
Before you start
- Docker with NVIDIA container support, as referenced by the nvidia-docker command
- A COCO data directory and a code directory to mount in the container
- Python and the NVIDIA PyTorch 20.06-py3 container image
Steps
Create the container using the README command, mounting the COCO data directory at /coco/ and the code directory at /yolo.
Clone the mish-cuda repository inside the container.
Run the command below in the mish-cuda directory.
python setup.py build installChange to the mounted code directory at /yolo.
Expected outcome
The documented container setup is prepared with mish-cuda installed and the code directory available at /yolo.
Guide limitations
- These steps cover setup only; they do not establish that the environment or commands have been independently tested.
- The README notes that a different PyTorch version may require trying another mish-cuda implementation.
Run the documented COCO model tests
Evaluate the P5, P6 or P7 weights using the repository’s COCO testing commands.
Before you start
- The setup described in the README
- COCO data configured through coco.yaml
- The corresponding yolov4-p5.pt, yolov4-p6.pt or yolov4-p7.pt weight file placed in /yolo/weights/
Steps
Download the weight file for the model variant and place it in /yolo/weights/.
Run the command below for P5, or use the corresponding documented P6 or P7 command.
python test.py --img 896 --conf 0.001 --batch 8 --device 0 --data coco.yaml --weights weights/yolov4-p5.ptReview the Average Precision and Average Recall metrics printed by the test.
Expected outcome
The README’s testing workflow reports object-detection evaluation metrics for the selected weights on the configured COCO data.
Guide limitations
- The documented commands specify device 0 and particular image sizes, confidence threshold and batch size.
- Reported repository results are not a guarantee of performance on other data or environments.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 14 October 2026
- Original website title
- WongKinYiu/ScaledYOLOv4 · Official homepage title · retained in its source language