Inference made using detect.py differs than inference made using PyTorch Hub inference

[edwardwang1]

Search before asking

• I have searched the YOLOv5 issues and discussions and found no similar questions.

Question

I have noticed that the inference obtained through the detect.py script differs than the inference made using PyTorch Hub. At first, I thought the issue was due to a discrepancy between the model downloaded from PyTorch Hub and the local copy of the model. However, the issue persisted even when loading the local model. The model weights were generated using the train.py script. Essentially, the bounding box is tighter in the PyTorch Hub case than the detect.py case.

I found a similar issue here. #2224

Inference via detect.py
!python detect.py --img-size 288 --conf 0.01 --project /content/saveDir2/ --save-conf --save-txt --source /content/temptest/ --weights "./yolo50epoch288V2_best.pt"
results = model(image).pandas().xywhn[0]
Results (x_cent, y_cent, width, height, conf) = (0.13 0.43 0.25 0.27 0.80)

Inference via PyTorch Hub
model = torch.hub.load('ultralytics/yolov5', 'custom', path="./yolo50epoch288V2_best.pt", force_reload=True)
results = model(image).pandas().xywhn[0]
Inference via PyTorch Hub (Local)
model = torch.hub.load('.', 'custom', path="./yolo50epoch288V2_best.pt", source="local", force_reload=True)
results = model(image).pandas().xywhn[0]
Results (x_cent, y_cent, width, height, conf) = (0.13, 0.43, 0.15, 0.14, 0.59)

Environment
I am running everything in a Google Colab Environment
YOLOv5 🚀 2022-1-22 torch 1.10.0+cu111 CUDA:0 (Tesla T4, 15110MiB)
YOLOv5 install
!git clone https://github.com/ultralytics/yolov5.git
!mv ./yolov5/* ./
!pip install -r requirements.txt

The issue also persists on my local machine.
YOLOv5 v6.0-205-gbd815d4 torch 1.9.0 CUDA:0 (NVIDIA GeForce RTX 3080, 10240MiB)

Additional

No response

[github-actions]

👋 Hello @edwardwang1, 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 screenshots and minimum viable code to reproduce your issue, otherwise we can not help you.

If this is a custom training ❓ Question, please provide as much information as possible, including dataset images, training logs, screenshots, and a public link to online W&B logging if available.

For business inquiries or professional support requests please visit https://ultralytics.com or email Glenn Jocher at glenn.jocher@ultralytics.com.

Requirements

Python>=3.6.0 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):

• Google Colab and Kaggle 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 (train.py), validation (val.py), inference (detect.py) and export (export.py) on MacOS, Windows, and Ubuntu every 24 hours and on every commit.

[glenn-jocher]

@edwardwang1 👋 hi, thanks for letting us know about this possible problem with YOLOv5 🚀. Your issue is not reproducible. Inference with detect.py and PyTorch Hub respectively for bus.jpg are below. If you are comparing both you should ensure all parameters are identical including batch size.

We've created a few short guidelines below to help users provide what we need in order to get started investigating a possible problem.

How to create a Minimal, Reproducible Example

When asking a question, people will be better able to provide help if you provide code that they can easily understand and use to reproduce the problem. This is referred to by community members as creating a minimum reproducible example. Your code that reproduces the problem should be:

• ✅ Minimal – Use as little code as possible to produce the problem
• ✅ Complete – Provide all parts someone else needs to reproduce the problem
• ✅ Reproducible – Test the code you're about to provide to make sure it reproduces the problem

For Ultralytics to provide assistance your code should also be:

• ✅ Current – Verify that your code is up-to-date with GitHub master, and if necessary git pull or git clone a new copy to ensure your problem has not already been solved in master.
• ✅ Unmodified – Your problem must be reproducible using official YOLOv5 code without changes. Ultralytics does not provide support for custom code ⚠️.

If you believe your problem meets all the above criteria, please close this issue and raise a new one using the 🐛 Bug Report template with a minimum reproducible example to help us better understand and diagnose your problem.

Thank you! 😃

[edwardwang1]

Hi Glenn,

Thanks for the prompt response. I tried the bus example using default yolov5x weights myself, and indeed found no discrepancy. It appears the issue is with my custom weights and custom classes.

I will update if I am able to resolve this.

[edwardwang1]

I've determined that my issue was due to not specifying the image size when using PyTorch Hub inference.
Wrong
results = model(im,).pandas().xywhn[0]
Correct
results = model(im, size=288).pandas().xywhn[0]

I previously didn't include the size flag because I already resize my images to 288x288 prior to inference.
Thanks again to Glenn for putting me on the right track.