52d9d99d4c33b94706cd392c44a703afcd69fa77,art/estimators/object_detection/pytorch_faster_rcnn.py,PyTorchFasterRCNN,loss_gradient,#PyTorchFasterRCNN#Any#Any#,138

Before Change


        loss.backward(retain_graph=True)  // type: ignore

        grad_list = list()
        for img in image_tensor_list:
            gradients = img.grad.cpu().numpy().copy()
            grad_list.append(gradients)

        grads = np.stack(grad_list, axis=0)

        // BB
        grads = self._apply_preprocessing_gradient(x, grads)

After Change


            grads = np.swapaxes(grads, 1, 3)
            grads = np.swapaxes(grads, 1, 2)
        else:
            grads = grads.permute(0, 3, 1, 2)

        assert grads.shape == x.shape

        if self.clip_values is not None:
Italian Trulli
In pattern: SUPERPATTERN

Frequency: 3

Non-data size: 2

Instances


Project Name: IBM/adversarial-robustness-toolbox
Commit Name: 52d9d99d4c33b94706cd392c44a703afcd69fa77
Time: 2020-11-11
Author: beat.buesser@ie.ibm.com
File Name: art/estimators/object_detection/pytorch_faster_rcnn.py
Class Name: PyTorchFasterRCNN
Method Name: loss_gradient


Project Name: Zhaoyi-Yan/Shift-Net_pytorch
Commit Name: 1483ec2a5bb7396d73339af3f07417e0c44b7965
Time: 2019-06-03
Author: yanzhaoyi@outlook.com
File Name: models/shift_net/InnerShiftTripleFunction.py
Class Name: InnerShiftTripleFunction
Method Name: backward


Project Name: yhenon/pytorch-retinanet
Commit Name: 83dc25dbc8b4ed98b4cc952e3cb35668e5a65490
Time: 2018-06-12
Author: yannhenon@gmail.com
File Name: coco_eval.py
Class Name:
Method Name: evaluate_coco