Warning! Only works for un-batched pointclouds. If they come batched we need to iterate over the batch dimension or do something REAL

Forward pass. :param predictions: <torch.FloatTensor: n_samples, C, H, W>. Predicted scores. :param targets: <torch.LongTenso

Calculate the loss. :param predictions: <torch.FloatTensor: n_samples, n_class, H, W>. Predicted scores. :param targets: <tor

Input: - pred: the output from model (before softmax), shape (N, C, H, W) - gt: ground truth map, shape (N, H, w)

Arguments: - strategy: RangeAug strategies. - image: original image, size: [bs, 6, h, w]. - label: origin

Arguments: - strategy: MixTeacher strategies. - image: original image, size: [6, H, W]. - label: original

Args: target: points gt labels for one batch xyz: coords of all points -> voxelized xyz : lidar.C[:, :3].float() or l

L_exp = w_dice * L_dice + w_cross * L_cross L_dice = E[(-ln(Dice))^{\gamma_{dice}}] E -> mean value Dice_i = [2(pred_score

To support a custom dataset, implement this function to receive the predicted results from the model, and then transform the unified

Args: pred_bbox3d: (N, 7) float Tensor. gt_bbox3d: (N, 7) float Tensor. Returns: corner_loss: (N) float Tensor.

Get padding parameters for np.pad function Args: desired_size: int, Desired padded output size cur_size: int, Current size. S

Multi-class Hinge Jaccard loss probas: [B, C, H, W] Variable, class probabilities at each prediction (between 0 and 1). Inter

modified from https://github.com/open-mmlab/mmdetection Args: tcp_port: backend: Returns:

Array of IoU for each (non ignored) class

IoU for foreground class binary: 1 foreground, 0 background

Something wrong with this loss Multi-class Lovasz-Softmax loss probas: [B, C, H, W] Variable, class probabilities at each prediction (b

Binary Lovasz hinge loss logits: [B, H, W] Variable, logits at each pixel (between -\infty and +\infty) labels: [B, H, W] Tensor, bin

Multi-class Lovasz-Softmax loss probas: [B, C, H, W] Variable, class probabilities at each prediction (between 0 and 1). Inte

Refer to https://github.com/tianweiy/CenterPoint. Modified focal loss. Exactly the same as CornerNet. Runs faster and costs a little bit more