012daf57ac31d82366323e7ec27ab7c2cb678a09,utils.py,,non_max_suppression,#Any#Any#Any#Any#,41

Before Change

            max_detections = torch.cat(max_detections)
            // Get index of image
            image_index = max_detections.new(max_detections.size(0), 1).fill_(image_i)
            // Repeat the batch_id for as many detections of the class cls in the image
            max_detections = torch.cat((image_index, max_detections), 1)
            // Add max detections to outputs

After Change

    box_corner[:, :, 3] = prediction[:, :, 1] + prediction[:, :, 3] / 2
    prediction[:, :, :4] = box_corner[:, :, :4]

    output = [None for _ in range(len(prediction))]
    for image_i, image_pred in enumerate(prediction):
        // Filter out confidence scores below threshold
        conf_mask = (image_pred[:, 4] >= conf_thres).squeeze()
        image_pred = image_pred[conf_mask]

Instances