Model output changes depending on whether Autoshape is used or not

[ymerkli]

Search before asking

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

YOLOv5 Component

AutoShape, DetectMultiBackend

Bug

In PR #9363, the following line was added to the constructor of the AutoShape wrapper:

yolov5/models/common.py

Line 603 in 23701ea

m.export = True # do not output loss values

i.e. if the model which AutoShape wraps is a PyTorch model, then the export class attribute of the Detect layer is set to True.

Why was this added and is this correct?

This now changes the behavior of the Detect layer depending on whether we use the AutoShape wrapper or not. The forward pass of the Detect layer returns tuple (output, input) if export=False and output if export=True. Thus, depending on whether we use AutoShape or DetectMultiBackend, the Detect layer (and thus the model) either outputs 1 or 2 values, which IMO it should not, the API of the Detect layer should not depend on the wrapper (which the Detect layer is not aware of). See the example below.

Environment

• YOLO: latest master
• OS: Ubuntu 20.04
• Python: 3.8.10

Minimal Reproducible Example

import torch

model_autoshape = torch.hub.load(
    "ultralytics/yolov5",
    model="yolov5s",
    pretrained=True,
    force_reload=True,
    autoshape=True,
)
model_dmb = torch.hub.load(
    "ultralytics/yolov5",
    model="yolov5s",
    pretrained=True,
    force_reload=True,
    autoshape=False,
)

images = torch.zeros((2, 3, 640, 640)).to(model_dmb.device)

output_autoshape = model_autoshape(images)
output_dmb = model_dmb(images)

print("AutoShape: ", type(output_autoshape), output_autoshape.shape)
print("DetectMultiBackend: ", type(output_dmb), [type(o) for o in output_dmb])

This outputs:

AutoShape:  <class 'torch.Tensor'> torch.Size([2, 25200, 85])
DetectMultiBackend:  <class 'list'> [<class 'torch.Tensor'>, <class 'list'>]

i.e. AutoShape returns the output of the detect layer only (the bounding box proposals), while DetectMultiBackend returns a tuple consisting of the Detect layer output and input.

Additional

No response

Are you willing to submit a PR?

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

[github-actions]

👋 Hello @ymerkli, 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 support@ultralytics.com.

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):

• 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]

@ymerkli yes this is true, the outputs are different now. The change was made in #9341 because the tuple outputs were not JIT traceable, which is a common use case, and so the current implementation use export=True to make them JIT traceable.

DetectMultiBackend is used by val.py and detect.py. When used by val.py the second output can be used to compute validation losses, but in practice is not used and only metrics are computed.

We could move the export flag setting to DMB, but then val.py will break as it's expecting 2 outputs always from the model (inference outputs and training outputs).

[github-actions]

👋 Hello, this issue has been automatically marked as stale because it has not had recent activity. Please note it will be closed if no further activity occurs.

Access additional YOLOv5 🚀 resources:

• Wiki – https://github.com/ultralytics/yolov5/wiki
• Tutorials – https://docs.ultralytics.com/yolov5
• Docs – https://docs.ultralytics.com

Access additional Ultralytics ⚡ resources:

• Ultralytics HUB – https://ultralytics.com/hub
• Vision API – https://ultralytics.com/yolov5
• About Us – https://ultralytics.com/about
• Join Our Team – https://ultralytics.com/work
• Contact Us – https://ultralytics.com/contact

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 YOLOv5 🚀 and Vision AI ⭐!