AttributeError: 'YoloHead' object has no attribute 'anchors'

[gsrujana]

🐛 Describe the bug

Getting this error when trying to fine-tune yolox model on version 3.2.0. This is issue with any of the yolox - l,s etc. No issues on using version 3.1.3.
[2023-08-30 21:02:03] INFO - checkpoint_utils.py - Successfully loaded pretrained weights for architecture yolox_l Traceback (most recent call last): File "/home/ubuntu/EFSDrive/Users/sgattupa/workspace/cxrs_greeniq/third_party/training_yoloX/train.py", line 40, in <module> model = models.get(MODEL_ARCH, num_classes=107, pretrained_weights="coco").to(DEVICE) File "/opt/conda/envs/pytorch/lib/python3.9/site-packages/super_gradients/training/models/model_factory.py", line 222, in get net = instantiate_model(model_name, arch_params, checkpoint_num_classes, pretrained_weights, download_required_code) File "/opt/conda/envs/pytorch/lib/python3.9/site-packages/super_gradients/training/models/model_factory.py", line 162, in instantiate_model net.replace_head(new_num_classes=num_classes_new_head) File "/opt/conda/envs/pytorch/lib/python3.9/site-packages/super_gradients/training/models/detection_models/yolo_base.py", line 684, in replace_head stride=self._head.anchors.stride, File "/opt/conda/envs/pytorch/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1207, in __getattr__ raise AttributeError("'{}' object has no attribute '{}'".format( AttributeError: 'YoloHead' object has no attribute 'anchors'

Versions

3.2.0

[Louis-Dupont]

Hi @gsrujana ,
This is a known issue that was recently addressed in this PR #1411
It will be part of the next release 3.2.1 which should come pretty soon

[gsrujana]

I see. For now, I added “self.anchors = anchors” in ln 413 of yolo_base.py (version 3.2.0) and training starts.

[BloodAxe]

Fixed in 3.2.1 release.