def post_process_inv_depth(inv_depth, inv_depth_flipped, method='mean'):
"""
Post-process an inverse and flipped inverse depth map
Parameters
----------
inv_depth : torch.Tensor [B,1,H,W]
Inverse depth map
inv_depth_flipped : torch.Tensor [B,1,H,W]
Inverse depth map produced from a flipped image
method : str
Method that will be used to fuse the inverse depth maps
Returns
-------
inv_depth_pp : torch.Tensor [B,1,H,W]
Post-processed inverse depth map
"""
B, C, H, W = inv_depth.shape
inv_depth_hat = flip_lr(inv_depth_flipped)
inv_depth_fused = fuse_inv_depth(inv_depth, inv_depth_hat, method=method)
xs = torch.linspace(0., 1., W, device=inv_depth.device,
dtype=inv_depth.dtype).repeat(B, C, H, 1)
mask = 1.0 - torch.clamp(20. * (xs - 0.05), 0., 1.)
mask_hat = flip_lr(mask)
return mask_hat * inv_depth + mask * inv_depth_hat + \
(1.0 - mask - mask_hat) * inv_depth_fused
def evaluate_depth(self, batch):
"""Evaluate batch to produce depth metrics."""
# Get predicted depth
inv_depths = self.model(batch)['inv_depths']
depth = inv2depth(inv_depths[0])
# Post-process predicted depth
batch['rgb'] = flip_lr(batch['rgb'])
inv_depths_flipped = self.model(batch)['inv_depths']
inv_depth_pp = post_process_inv_depth(inv_depths[0],
inv_depths_flipped[0],
method='mean')
depth_pp = inv2depth(inv_depth_pp)
batch['rgb'] = flip_lr(batch['rgb'])
# Calculate predicted metrics
metrics = OrderedDict()
if 'depth' in batch:
for mode in self.metrics_modes:
if self.config.model.loss.mask_ego:
metrics[self.metrics_name +
mode] = compute_ego_depth_metrics(
self.config.model.params,
gt=batch['depth'],
pred=depth_pp if 'pp' in mode else depth,
path_to_ego_masks=batch['path_to_ego_mask'],
use_gt_scale='gt' in mode)
else:
metrics[self.metrics_name + mode] = compute_depth_metrics(
self.config.model.params,
gt=batch['depth'],
pred=depth_pp if 'pp' in mode else depth,
use_gt_scale='gt' in mode)
# Return metrics and extra information
return {'metrics': metrics, 'inv_depth': inv_depth_pp}
def evaluate_depth(self, batch):
"""Evaluate batch to produce depth metrics."""
# Get predicted depth
inv_depths = self.model(batch)["inv_depths"]
save_image(inv_depths[0], "inv_depths.png") # new
depth = inv2depth(inv_depths[0])
# Post-process predicted depth
batch["rgb"] = flip_lr(batch["rgb"])
inv_depths_flipped = self.model(batch)["inv_depths"]
inv_depth_pp = post_process_inv_depth(inv_depths[0],
inv_depths_flipped[0],
method="mean")
depth_pp = inv2depth(inv_depth_pp)
batch["rgb"] = flip_lr(batch["rgb"])
# Calculate predicted metrics
metrics = OrderedDict()
if "depth" in batch:
for mode in self.metrics_modes:
metrics[self.metrics_name + mode] = compute_depth_metrics(
self.config.model.params,
gt=batch["depth"],
pred=depth_pp if "pp" in mode else depth,
use_gt_scale="gt" in mode,
)
# Return metrics and extra information
return {"metrics": metrics, "inv_depth": inv_depth_pp}