#3939th of 70 in Open models
d-fine
Real-time DETR object detector with COCO and Objects365 checkpoints
- Stars
- 3.4k
- License
- Apache-2.0
- Last commit
- Aug 2026
- Language
- Python
3,057 on 21 Apr 20263,400 on 10 Nov 2026
From GitHub's daily star counts.
Checked on GitHub.
Overview
Official PyTorch implementation of D-FINE, a real-time DETR-style object detector that treats box regression as fine-grained distribution refinement with self-distillation. It ships five sizes (N to X, 4M to 62M parameters) with training, testing and fine-tuning configs for COCO, Objects365 and custom COCO-format datasets. The N model reports 42.8 AP on COCO and the X model 55.8 AP, or 59.3 AP with Objects365 pretraining.
Who it is for: Computer vision engineers training or fine-tuning real-time object detectors
Strengths
- Five model sizes, N to X, with COCO AP from 42.8 to 55.8
- Latency listed per model on a T4 with TensorRT 10.4.0, fp16
- Configs and commands for fine-tuning on custom COCO-format datasets
- Checkpoints and training logs published for every listed model
Weaknesses
- Objects365-trained checkpoints may not be cleared for commercial use
- Training commands assume a 4-GPU torchrun setup
- Object detection only, no segmentation or language features
- Objects365 data prep needs several manual remap and resize steps
What it needs
- GPU required
- Docker
- Needs PyTorch, conda, Python 3.11.9, TensorRT (for latency benchmarks)
- Models: D-FINE N/S/M/L/X (HGNetv2), COCO, Objects365, custom COCO-format datasets
Quick answers
Is d-fine open source?
Yes. Its code is on GitHub under the Apache-2.0 license.
Can you run d-fine with Docker?
Yes. Its repository has a Docker setup: a Dockerfile, a Compose file or a published image.
Does d-fine need a GPU?
Yes, it needs a GPU.
Also in Open models
See all 70| Rank | Project | Score |
|---|---|---|
| 1 | ruviewWiFi CSI sensing on ESP32 for presence, breathing, heart rate and pose | 94 out of 100 |
| 2 | layaSingle-pass multilingual classifier for typed choice, score and yes/no decisions | 88 out of 100 |
| 3 | rf-detrReal-time transformer models for object detection, segmentation and keypoints | 79 out of 100 |
| #4 | timesfmPretrained time-series foundation model for zero-shot forecasting from Google Research | 78 out of 100 |
| #5 | yolov5PyTorch YOLOv5 object detection, segmentation and classification models with training code | 76 out of 100 |