NVIDIA / NVIDIA/TensorRT

YOLOv10 Error with TensorRT 10.13

Open
#4,820 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Module:Accuracy
Dominant language
C++
Stars
13.4k
Forks
2.4k
Avg merge
5d 3h
Merged PRs (30d)
2

Description

With TensorRT 10.3(R36.4.3), the inference results of both FP32 and FP16 are close to ONNX Runtime. However, after using TensorRT 10.13(R38.4.0), only FP32 inference remains consistent with ONNX Runtime, while FP16 shows significant discrepancies.

Reproduction steps:
python3 export_onnx.py
python3 run.py

export_onnx.py generates yolov10n.onnx, which can be used for the following tests.
run.py generates three output images: ort_result.png, trt_result.png, and trt_fp32_result.png.

Note: export_onnx.py was tested with Ultralytics 8.4.98 and PyTorch 2.8.0+cu126.
file.zip

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start by running export_onnx.py and run.py with the provided file.zip and compare ort_result.png, trt_result.png, and trt_fp32_result.png. Investigate the FP16 discrepancy between TensorRT 10.13 and ONNX Runtime while confirming FP32 remains consistent; done means the FP16 output matches the expected inference results.

Written by the indexing model from the issue text.

Assessment

Tech stack
python, pytorch
Domain
computer-vision, machine-learning, performance
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Quiet
Clarity
Mostly clear
Newbie friendliness
42/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.