YOLOv10: Real-Time End-to-End Object Detection
Authors: Ao Wang, Hui Chen, Lihao Liu, Kai CHEN, Zijia Lin, Jungong Han, guiguang ding
NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments show that YOLOv10 achieves the state-of-the-art performance and efficiency across various model scales. For example, our YOLOv10-S is 1.8 faster than RT-DETR-R18 under the similar AP on COCO, meanwhile enjoying 2.8 smaller number of parameters and FLOPs. Compared with YOLOv9-C, YOLOv10-B has 46% less latency and 25% fewer parameters for the same performance. |
| Researcher Affiliation | Academia | Ao Wang1 Hui Chen2 Lihao Liu1 Kai Chen1 Zijia Lin1 Jungong Han3 Guiguang Ding1 1School of Software, Tsinghua University 2BNRist, Tsinghua University 3Department of Automation, Tsinghua University |
| Pseudocode | Yes | Algorithm 1: Rank-guided block design |
| Open Source Code | Yes | Code and models are available at https://github.com/THU-MIG/yolov10. |
| Open Datasets | Yes | Extensive experiments on standard benchmarks for object detection, i.e., COCO [35], demonstrate that our YOLOv10 can significantly outperform previous state-of-the-art models in terms of computation-accuracy trade-offs across various model scales. [...] the data is available at https://www.cocodataset.org/. |
| Dataset Splits | Yes | We verify the proposed detector on COCO [35] under the same training-from-scratch setting [21, 65, 62]. Moreover, the latencies of all models are tested on T4 GPU with Tensor RT FP16, following [78]. ... We report the standard mean average precision (AP) across different object scales and Io U thresholds on the COCO dataset [35]. |
| Hardware Specification | Yes | All models are trained on 8 NVIDIA 3090 GPUs. ...the latencies of all models are tested on T4 GPU with Tensor RT FP16... ...on CPU (Intel Xeon Skylake, IBRS) using Open VINO |
| Software Dependencies | No | The paper mentions 'Tensor RT FP16' and 'Open VINO' but does not specify version numbers for these or other relevant software libraries (e.g., PyTorch, TensorFlow, CUDA). |
| Experiment Setup | Yes | We employ the consistent dual assignments for NMS-free training and perform holistic efficiency-accuracy driven model design based on it, which brings our YOLOv10 models. ... We verify the proposed detector on COCO [35] under the same training-from-scratch setting [21, 65, 62]. ... Tab. 14 presents the detailed hyper-parameters. hyper-parameter YOLOv10-N/S/M/B/L/X epochs 500 optimizer SGD momentum 0.937 weight decay 5 10 4 warm-up epochs 3 ... initial learning rate 10 2 final learning rate 10 4 |