src/models/tools/losses.py

import torch
import torch.nn.functional as F
from .hessian_penalty import hessian_penalty
from .mmd import compute_mmd


def compute_rc_loss(model, batch, use_txt_output=False):
    x = batch["x"]
    output = batch["output"]
    mask = batch["mask"]
    if use_txt_output:
        output = batch["txt_output"]
    gtmasked = x.permute(0, 3, 1, 2)[mask]
    outmasked = output.permute(0, 3, 1, 2)[mask]

    loss = F.mse_loss(gtmasked, outmasked, reduction='mean')
    return loss


def compute_rcxyz_loss(model, batch, use_txt_output=False):
    x = batch["x_xyz"]
    output = batch["output_xyz"]
    mask = batch["mask"]
    if use_txt_output:
        output = batch["txt_output_xyz"]
    gtmasked = x.permute(0, 3, 1, 2)[mask]
    outmasked = output.permute(0, 3, 1, 2)[mask]

    loss = F.mse_loss(gtmasked, outmasked, reduction='mean')
    return loss


def compute_vel_loss(model, batch, use_txt_output=False):
    x = batch["x"]
    output = batch["output"]
    if use_txt_output:
        output = batch["txt_output"]
    gtvel = (x[..., 1:] - x[..., :-1])
    outputvel = (output[..., 1:] - output[..., :-1])

    mask = batch["mask"][..., 1:]

    gtvelmasked = gtvel.permute(0, 3, 1, 2)[mask]
    outvelmasked = outputvel.permute(0, 3, 1, 2)[mask]

    loss = F.mse_loss(gtvelmasked, outvelmasked, reduction='mean')
    return loss


def compute_velxyz_loss(model, batch, use_txt_output=False):
    x = batch["x_xyz"]
    output = batch["output_xyz"]
    if use_txt_output:
        output = batch["txt_output_xyz"]
    gtvel = (x[..., 1:] - x[..., :-1])
    outputvel = (output[..., 1:] - output[..., :-1])

    mask = batch["mask"][..., 1:]

    gtvelmasked = gtvel.permute(0, 3, 1, 2)[mask]
    outvelmasked = outputvel.permute(0, 3, 1, 2)[mask]

    loss = F.mse_loss(gtvelmasked, outvelmasked, reduction='mean')
    return loss


def compute_hp_loss(model, batch):
    loss = hessian_penalty(model.return_latent, batch, seed=torch.random.seed())
    return loss


def compute_mmd_loss(model, batch):
    z = batch["z"]
    true_samples = torch.randn(z.shape, requires_grad=False, device=model.device)
    loss = compute_mmd(true_samples, z)
    return loss


_matching_ = {"rc": compute_rc_loss, "hp": compute_hp_loss,
              "mmd": compute_mmd_loss, "rcxyz": compute_rcxyz_loss,
              "vel": compute_vel_loss, "velxyz": compute_velxyz_loss}


def get_loss_function(ltype):
    return _matching_[ltype]


def get_loss_names():
    return list(_matching_.keys())