e583ae1925021107a0672e5cd4f586aef3e50452,pytorch/seg_metrics.py,PixelAccuracyMetric,update,#PixelAccuracyMetric#Any#Any#,47

Before Change

        
        check_label_shapes(labels, preds)
        for label, pred in zip(labels, preds):
            pred = pred.max(dim=self.axis)[1]
            label = label.max(dim=self.axis)[1]
            self.sum_metric += pred.eq(label).sum()
            self.num_inst += pred.numel()

    def reset(self):
        

After Change

            Predicted values.
        
        check_label_shapes(labels, preds)
        if self.on_cpu:
            if self.sparse_label:
                label_imask = labels.cpu().numpy().astype(np.int32)
            else:
                label_imask = torch.argmax(labels, dim=self.axis).cpu().numpy().astype(np.int32)
            pred_imask = torch.argmax(preds, dim=self.axis).cpu().numpy().astype(np.int32)
            acc = seg_pixel_accuracy_np(
                label_imask=label_imask,
                pred_imask=pred_imask,
                vague_idx=self.vague_idx,
                use_vague=self.use_vague,
                macro_average=self.macro_average)
            if self.macro_average:
                self.sum_metric += acc
                self.num_inst += 1
            else:
                self.sum_metric += acc[0]
                self.num_inst += acc[1]
        else:
            assert False

    def reset(self):
        
        Resets the internal evaluation result to initial state.
        

Instances