Metrics

This section introduce the metrics included in Splatwizard for training and evaluation.

• splatwizard.metrics.l1_func(img1: Tensor, img2: Tensor)

  Calculate L1 distance, available for training and evaluation.

• splatwizard.metrics.l2_func(img1: Tensor, img2: Tensor)

  Calculate L2 distance, available for training and evaluation.

• splatwizard.metrics.union_ssim_func(img1: Tensor, img2: Tensor, using_fused=True)

  Calculate SSIM value, available for training. The function uses fussed_ssim lib when specify using_fused=True.

• splatwizard.metrics.ssim_func(img1: Tensor, img2: Tensor, window_size=11, size_average=True)

  Calculate SSIM value, available for training and evaluation. The batch dimension will be kept if size_average=False

• splatwizard.metrics.lpips_func(img1: Tensor, img2: Tensor, ret_per_layer=False, normalization=True)

  Calculate LPIPS(VGG) value, recommended for evaluation only. The batch dimension will be kept.

• splatwizard.metrics.mse_func(img1: Tensor, img2: Tensor)

  Calculate mean squared error value, recommended for evaluation only. The batch dimension will be kept.

• splatwizard.metrics.psnr_func(img1: Tensor, img2: Tensor)

  Calculate peak-signal-noise-ratio value, recommended for evaluation only. The batch dimension will be kept.