DL_notes.txt

class net ()

	forward propagation 

	calculation of loss

	backward propagation

defining  the forward fucntion first and backward fucntion is automatically defined using backprop

any of the tensor operations can be used in forward pass



Point-wise Mutual Information (PMI) loss: This loss function is based on the mutual information between the predicted
 segmentation and the input point cloud. It encourages the predicted segmentation to be consistent with the input point
  cloud by penalizing the mutual information between the predicted segmentation and the input point cloud.

Focal loss: This loss function is designed to address class imbalance in the training data by assigning higher weights
 to misclassified examples. It is commonly used as a loss function for point cloud semantic segmentation tasks.

Dice loss: This loss function is based on the Dice coefficient, which measures the overlap between the predicted
 segmentation and the ground truth segmentation. It is commonly used as a loss function for binary segmentation
  tasks, but can be adapted for multi-class segmentation tasks.

Boundary loss: This loss function encourages the predicted segmentation to have sharp and accurate boundaries by
 penalizing the distance between the predicted segmentation and the ground truth segmentation along the boundary.

Lovasz-Softmax loss: This loss function is based on the Lovasz extension of the submodular set function, which measures
 the difference between the predicted segmentation and the ground truth segmentation in terms of the order of the
  predicted labels. It is commonly used as a loss function for multi-class segmentation tasks.

Chamfer distance: This loss function measures the distance between two point clouds by computing the minimum distance
 between each point in one cloud and its nearest neighbor in the other cloud. It is commonly used as a loss function
  for point cloud registration and reconstruction tasks.

Earth Mover's Distance (EMD): This loss function is similar to the Chamfer distance, but instead of computing the
 minimum distance between each point and its nearest neighbor, it computes the minimum cost of transforming one point 
 cloud into the other by moving each point a certain distance. It is commonly used as a loss function for point cloud
  generation tasks.

Repulsion loss: This loss function encourages the points in a point cloud to be evenly distributed by penalizing the
 distance between nearby points. It is commonly used as a regularization term in conjunction with other loss functions.

F-score loss: This loss function is based on the F-score metric, which measures the harmonic mean of precision and
 recall. It is commonly used as a loss function for point cloud segmentation tasks.