KeyError: 'model.24.anchor_grid' with v6.0

[deepsworld]

❔Question

There is an additional key called model.24.anchor_grid in yolov5-v6.0 checkpoints which does not exist in the model loaded with hub.load(pretrained=False). I am not sure if this is intended or I am missing any argument. I appreciate your help.

Additional context

Minimal code to reproduce the error

import torch
from torch import hub
# pytorch 1.9 workaround 
torch.hub._validate_not_a_forked_repo=lambda a,b,c: True

def github_release_url(owner, project, tag, name):
    return f"https://github.com/{owner}/{project}/releases/download/{tag}/{name}"

tag = 'v6.0'
chkpt = 'yolov5x'
model = hub.load(f'ultralytics/yolov5:{tag}', chkpt, False, 3, 80, force_reload=True, autoshape=False)
url = github_release_url('ultralytics', 'yolov5', tag, f'{chkpt}.pt')
chkpt = hub.load_state_dict_from_url(url,
                                     model_dir=None,
                                     map_location=torch.device('cpu'))
state_dict = chkpt['model'].float().state_dict()

try:
    state_dict = { k: v for k, v in state_dict.items() if model.state_dict()[k].shape == v.shape }
except Exception as e:
    raise e

[github-actions]

👋 Hello @deepsworld, 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
$ cd yolov5
$ pip install -r requirements.txt

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.

[deepsworld]

NOTE: The version v5.0 when loaded in similar fashion works fine.
@glenn-jocher Is there a reason the register buffer for anchor grid was removed from the Detect module and only added when pretrained is True. Here: https://github.com/ultralytics/yolov5/blob/v5.0/models/yolo.py#L37

[glenn-jocher]

@deepsworld yes this change was introduced in OpenCV DNN PR #4833, along with backward compatibility updates to allow previous version models to load and run correctly.

[Wangwang99999]

I don't see any answer to this question. Why there is no 'model.24.anchor_grid' in trained model, but it do exist in pretrained models.

[glenn-jocher]

@Wangwang99999 architecture may update over time