[yolov5s][INT8] Quantizing yolov5s model with batch-size > 1 to OpenVINO format

[txlim96]

Search before asking

• I have searched the YOLOv5 issues and found no similar bug report.

YOLOv5 Component

Export

Bug

When exporting the quantized yolov5s model with batch-size=1, the openvino IR files can be exported

However, when changing the batch-size to a different number, 4 in this scenario, the following error popped up

It seems to me it somehow is related to the batch size causing some issues in reshaping, however without quantizing (without --int8 argument), the model is able to be exported to openvino IR with different batch sizes, not sure is this a yolov5 repo issue or openvino issue?

Environment

• YOLO: YOLOv5 🚀 v7.0-193-g485da42 Python-3.10.6 torch-2.0.1+cu117 CPU
• Ubuntu: 22.04
• Python: 3.10.6

Minimal Reproducible Example

python3 export.py --weights yolov5s.pt --int8 --include openvino --batch-size 4

Additional

No response

Are you willing to submit a PR?

• Yes I'd like to help by submitting a PR!

[github-actions]

👋 Hello @txlim96, thank you for your interest in YOLOv5 🚀! Please visit our ⭐️ Tutorials to get started, where you can find quickstart guides for simple tasks like Custom Data Training all the way to advanced concepts like Hyperparameter Evolution.

If this is a 🐛 Bug Report, please provide a minimum reproducible example to help us debug it.

If this is a custom training ❓ Question, please provide as much information as possible, including dataset image examples and training logs, and verify you are following our Tips for Best Training Results.

Requirements

Python>=3.7.0 with all requirements.txt installed including PyTorch>=1.7. To get started:

git clone https://github.com/ultralytics/yolov5  # clone
cd yolov5
pip install -r requirements.txt  # install

Environments

YOLOv5 may be run in any of the following up-to-date verified environments (with all dependencies including CUDA/CUDNN, Python and PyTorch preinstalled):

• Notebooks with free GPU:
• Google Cloud Deep Learning VM. See GCP Quickstart Guide
• Amazon Deep Learning AMI. See AWS Quickstart Guide
• Docker Image. See Docker Quickstart Guide

Status

If this badge is green, all YOLOv5 GitHub Actions Continuous Integration (CI) tests are currently passing. CI tests verify correct operation of YOLOv5 training, validation, inference, export and benchmarks on macOS, Windows, and Ubuntu every 24 hours and on every commit.

Introducing YOLOv8 🚀

We're excited to announce the launch of our latest state-of-the-art (SOTA) object detection model for 2023 - YOLOv8 🚀!

Designed to be fast, accurate, and easy to use, YOLOv8 is an ideal choice for a wide range of object detection, image segmentation and image classification tasks. With YOLOv8, you'll be able to quickly and accurately detect objects in real-time, streamline your workflows, and achieve new levels of accuracy in your projects.

Check out our YOLOv8 Docs for details and get started with:

pip install ultralytics

[glenn-jocher]

@txlim96 hi there! Thank you for bringing this issue to our attention. We appreciate your detailed report.

Based on the error you encountered when changing the batch size to a different number during the export process, it appears that there might be a reshaping issue related to the batch size. Without quantization (without the --int8 argument), you mentioned that the model can be successfully exported to openvino IR with different batch sizes.

To determine whether this is a YOLOv5 repository issue or an OpenVINO issue, it would be helpful to investigate further. Could you please provide some additional information?

1. Are you able to export a quantized model with batch size other than 1 using other formats apart from OpenVINO?
2. Have you tried running the quantized model with a batch size other than 1 in other frameworks (e.g., PyTorch) to assess if the error is specific to OpenVINO?

Please let us know your findings, and we will be happy to assist you further in resolving this issue. Thank you for your valuable contribution!

[txlim96]

Hi @glenn-jocher,

1. I did try out with other int8-available model formats like coreml and tflite, and they can be exported
2. I did run the quantized tflite model (batch-size=4) with val.py and looks like there is a dimension mismatch error

# Export model
python3 export.py --weights yolov5s.pt --int8 --include tflite --batch-size 4
# Validation
python3 val.py --weights yolov5s-int8.tflite --batch-size 4

[glenn-jocher]

Hi @txlim96,

Thank you for providing the additional information. Based on your findings, it seems that the issue lies with the model export process when using batch size other than 1 with quantization. The dimension mismatch error you encountered when running the quantized tflite model with a batch size of 4 further confirms this.

This suggests that there may be certain constraints or limitations when quantizing the model with a batch size other than 1 using YOLOv5's current implementation. It's possible that the reshaping operations during the quantization process are not handling the batch dimension correctly.

To address this issue, I recommend looking into the specific code responsible for the quantization process in YOLOv5 and checking for any potential issues related to reshaping operations with different batch sizes.

Alternatively, you may also consider reaching out to the YOLOv5 community or the Ultralytics team for further assistance. They have extensive knowledge and expertise with the YOLOv5 framework and may be able to provide more insights or guidance on this issue.

Thank you for bringing this to our attention, and, once again, thank you for your valuable contribution to the YOLOv5 project!

[github-actions]

👋 Hello there! We wanted to give you a friendly reminder that this issue has not had any recent activity and may be closed soon, but don't worry - you can always reopen it if needed. If you still have any questions or concerns, please feel free to let us know how we can help.

For additional resources and information, please see the links below:

• Docs: https://docs.ultralytics.com
• HUB: https://hub.ultralytics.com
• Community: https://community.ultralytics.com

Feel free to inform us of any other issues you discover or feature requests that come to mind in the future. Pull Requests (PRs) are also always welcomed!

Thank you for your contributions to YOLO 🚀 and Vision AI ⭐