#4444th of 70 in Open models
deimv2
Real-time object detector family built on DINOv3 and HGNetv2 backbones
- Stars
- 2.1k
- License
- custom license
- Last commit
- Oct 2026
- Language
- Jupyter Notebook
1,684 on 21 Apr 20262,099 on 10 Nov 2026
From GitHub's daily star counts.
Checked on GitHub.
Overview
DEIMv2 is a PyTorch object detection codebase with eight COCO-trained variants, from Atto (0.5M params, 23.8 AP) to X (50.3M params, 57.8 AP). The S, M, L and X models use DINOv3-derived backbones; the smaller ones use HGNetv2. It includes training, testing and fine-tuning via torchrun, custom COCO-format datasets, ONNX and TensorRT deployment tools, and Hugging Face checkpoints.
Who it is for: ML engineers training or deploying real-time object detectors on edge or GPU
Strengths
- Eight sizes, from Atto at 0.8 GFLOPs to X at 57.8 COCO AP
- S model reaches 50.9 AP with 9.7M params and 5.78 ms latency
- Checkpoints, configs and logs published, plus Hugging Face weights
- Custom datasets supported in COCO format via YAML config
Weaknesses
- Custom DEIMv2 License, not a standard OSI license
- Object detection only; pose and segmentation are in EdgeCrafter
- DINOv3 and distilled ViT backbones must be downloaded manually into ckpts/
- No Docker setup; documented training uses multi-GPU torchrun
What it needs
- GPU required
- Needs PyTorch 2.5.1 or 2.4.1, Python 3.11, DINOv3 checkpoints, onnx, onnxsim, TensorRT >= 10.6 for FP16
- Models: DEIMv2 Atto/Femto/Pico/N (HGNetv2), DEIMv2 S/M/L/X (DINOv3), COCO-format datasets, ONNX, TensorRT
Quick answers
Is deimv2 open source?
Its code is public on GitHub under its own license terms; read them before you use it at work.
Does deimv2 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 |