Adversarial Patch - PyTorch - YOLO notebook

[dalqattan]

Hi

I tried to implement Adversarial Patch - PyTorch - YOLO notebook here

I faced two issues

First: Model definition

When I run the Model definition code I got this error

WARNING ⚠️ DetectMultiBackend failed: cannot import name 'MobileOptimizerType' from 'torch._C' (/usr/local/lib/python3.10/dist-packages/torch/_C.cpython-310-x86_64-linux-gnu.so)
---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
[<ipython-input-10-ac0fa7c6b73a>](https://localhost:8080/#) in <cell line: 4>()
     74     model = yolov5.load('yolov5s.pt')
     75 
---> 76     model = Yolo(model)
     77 
     78     detector = PyTorchYolo(model=model,

1 frames
[<ipython-input-10-ac0fa7c6b73a>](https://localhost:8080/#) in __init__(self, model)
     56                             'fl_gamma': 0.0
     57                             }
---> 58             self.compute_loss = ComputeLoss(self.model.model.model)
     59 
     60         def forward(self, x, targets=None):

[/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py](https://localhost:8080/#) in __getattr__(self, name)
   1612         for module_prefix, module in modules:
   1613             members = get_members_fn(module)
-> 1614             for k, v in members:
   1615                 if v is None or v in memo:
   1616                     continue

AttributeError: 'Sequential' object has no attribute 'model'

I change the line 58 to

self.compute_loss = ComputeLoss(self.model)
SO I got no error

However, I faced another error

Second: Detector Predictions on Benign Images

When I run the Detector Predictions on Benign Images code, I got this error

AttributeError                            Traceback (most recent call last)
[<ipython-input-9-9793e2716346>](https://localhost:8080/#) in <cell line: 2>()
      1 get_ipython().run_line_magic('matplotlib', 'inline')
----> 2 dets = detector.predict(coco_images)
      3 for i in range(len(dets)):
      4     preds_orig = extract_predictions(dets[i], 0.8)
      5     plot_image_with_boxes(img=coco_images[i].transpose(1,2,0).copy(), boxes=preds_orig[1], pred_cls=preds_orig[0], title="Predictions on image without patch")

1 frames
[/usr/local/lib/python3.10/dist-packages/art/estimators/object_detection/pytorch_yolo.py](https://localhost:8080/#) in predict(self, x, batch_size, **kwargs)
    463                 predictions_xcycwh = self._model(x_batch)
    464 
--> 465             predictions_x1y1x2y2 = translate_predictions_xcycwh_to_x1y1x2y2(
    466                 y_pred_xcycwh=predictions_xcycwh, height=height, width=width
    467             )

[/usr/local/lib/python3.10/dist-packages/art/estimators/object_detection/pytorch_yolo.py](https://localhost:8080/#) in translate_predictions_xcycwh_to_x1y1x2y2(y_pred_xcycwh, height, width)
     55 
     56     y_pred_x1y1x2y2 = []
---> 57     device = y_pred_xcycwh.device
     58 
     59     for y_pred in y_pred_xcycwh:

AttributeError: 'tuple' object has no attribute 'device'

Any help please.

Many Thanks

Duaa

[beat-buesser]

Hi @dalqattan Which version of Yolo are you using?

[dalqattan]

YOLOv5

[beat-buesser]

In the 4th cell of the notebook, what did you set for variable MODEL? Default is yolov3 but it probably should be yolov5.

[dalqattan]

yolov5