HongChuang/ultralytics-ascend
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 3

Add docformatter to pre-commit (#5279)

Browse source 
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Burhan <62214284+Burhan-Q@users.noreply.github.com>
This commit is contained in:
Glenn Jocher 2023-10-09 02:25:22 +02:00 • committed by GitHub
parent c7aa83da31
commit 7517667a33
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
90 changed files with 1396 additions and 497 deletions
Download patch file Download diff file Expand all files Collapse all files

43
ultralytics/models/utils/loss.py
View file

@ -47,6 +47,7 @@ class DETRLoss(nn.Module):
self.device = None
def _get_loss_class(self, pred_scores, targets, gt_scores, num_gts, postfix=''):
"""Computes the classification loss based on predictions, target values, and ground truth scores."""
# logits: [b, query, num_classes], gt_class: list[[n, 1]]
name_class = f'loss_class{postfix}'
bs, nq = pred_scores.shape[:2]
@ -68,6 +69,9 @@ class DETRLoss(nn.Module):
return {name_class: loss_cls.squeeze() * self.loss_gain['class']}
def _get_loss_bbox(self, pred_bboxes, gt_bboxes, postfix=''):
"""Calculates and returns the bounding box loss and GIoU loss for the predicted and ground truth bounding
boxes.
"""
# boxes: [b, query, 4], gt_bbox: list[[n, 4]]
name_bbox = f'loss_bbox{postfix}'
name_giou = f'loss_giou{postfix}'
@ -125,7 +129,7 @@ class DETRLoss(nn.Module):
postfix='',
masks=None,
gt_mask=None):
"""Get auxiliary losses"""
"""Get auxiliary losses."""
# NOTE: loss class, bbox, giou, mask, dice
loss = torch.zeros(5 if masks is not None else 3, device=pred_bboxes.device)
if match_indices is None and self.use_uni_match:
@ -166,12 +170,14 @@ class DETRLoss(nn.Module):
@staticmethod
def _get_index(match_indices):
"""Returns batch indices, source indices, and destination indices from provided match indices."""
batch_idx = torch.cat([torch.full_like(src, i) for i, (src, _) in enumerate(match_indices)])
src_idx = torch.cat([src for (src, _) in match_indices])
dst_idx = torch.cat([dst for (_, dst) in match_indices])
return (batch_idx, src_idx), dst_idx
def _get_assigned_bboxes(self, pred_bboxes, gt_bboxes, match_indices):
"""Assigns predicted bounding boxes to ground truth bounding boxes based on the match indices."""
pred_assigned = torch.cat([
t[I] if len(I) > 0 else torch.zeros(0, t.shape[-1], device=self.device)
for t, (I, _) in zip(pred_bboxes, match_indices)])
@ -190,7 +196,7 @@ class DETRLoss(nn.Module):
gt_mask=None,
postfix='',
match_indices=None):
"""Get losses"""
"""Get losses."""
if match_indices is None:
match_indices = self.matcher(pred_bboxes,
pred_scores,
@ -250,22 +256,43 @@ class DETRLoss(nn.Module):
class RTDETRDetectionLoss(DETRLoss):
"""
Real-Time DeepTracker (RT-DETR) Detection Loss class that extends the DETRLoss.
This class computes the detection loss for the RT-DETR model, which includes the standard detection loss as well as
an additional denoising training loss when provided with denoising metadata.
"""
def forward(self, preds, batch, dn_bboxes=None, dn_scores=None, dn_meta=None):
"""
Forward pass to compute the detection loss.
Args:
preds (tuple): Predicted bounding boxes and scores.
batch (dict): Batch data containing ground truth information.
dn_bboxes (torch.Tensor, optional): Denoising bounding boxes. Default is None.
dn_scores (torch.Tensor, optional): Denoising scores. Default is None.
dn_meta (dict, optional): Metadata for denoising. Default is None.
Returns:
(dict): Dictionary containing the total loss and, if applicable, the denoising loss.
"""
pred_bboxes, pred_scores = preds
total_loss = super().forward(pred_bboxes, pred_scores, batch)
# Check for denoising metadata to compute denoising training loss
if dn_meta is not None:
dn_pos_idx, dn_num_group = dn_meta['dn_pos_idx'], dn_meta['dn_num_group']
assert len(batch['gt_groups']) == len(dn_pos_idx)
# Denoising match indices
# Get the match indices for denoising
match_indices = self.get_dn_match_indices(dn_pos_idx, dn_num_group, batch['gt_groups'])
# Compute denoising training loss
# Compute the denoising training loss
dn_loss = super().forward(dn_bboxes, dn_scores, batch, postfix='_dn', match_indices=match_indices)
total_loss.update(dn_loss)
else:
# If no denoising metadata is provided, set denoising loss to zero
total_loss.update({f'{k}_dn': torch.tensor(0., device=self.device) for k in total_loss.keys()})
return total_loss
@ -276,12 +303,12 @@ class RTDETRDetectionLoss(DETRLoss):
Get the match indices for denoising.
Args:
dn_pos_idx (List[torch.Tensor]): A list includes positive indices of denoising.
dn_num_group (int): The number of groups of denoising.
gt_groups (List(int)): a list of batch size length includes the number of gts of each image.
dn_pos_idx (List[torch.Tensor]): List of tensors containing positive indices for denoising.
dn_num_group (int): Number of denoising groups.
gt_groups (List[int]): List of integers representing the number of ground truths for each image.
Returns:
dn_match_indices (List(tuple)): Matched indices.
(List[tuple]): List of tuples containing matched indices for denoising.
"""
dn_match_indices = []
idx_groups = torch.as_tensor([0, *gt_groups[:-1]]).cumsum_(0)

32
ultralytics/models/utils/ops.py
View file

@ -11,8 +11,8 @@ from ultralytics.utils.ops import xywh2xyxy, xyxy2xywh
class HungarianMatcher(nn.Module):
"""
A module implementing the HungarianMatcher, which is a differentiable module to solve the assignment problem in
an end-to-end fashion.
A module implementing the HungarianMatcher, which is a differentiable module to solve the assignment problem in an
end-to-end fashion.
HungarianMatcher performs optimal assignment over the predicted and ground truth bounding boxes using a cost
function that considers classification scores, bounding box coordinates, and optionally, mask predictions.
@ -32,6 +32,9 @@ class HungarianMatcher(nn.Module):
"""
def __init__(self, cost_gain=None, use_fl=True, with_mask=False, num_sample_points=12544, alpha=0.25, gamma=2.0):
"""Initializes HungarianMatcher with cost coefficients, Focal Loss, mask prediction, sample points, and alpha
gamma factors.
"""
super().__init__()
if cost_gain is None:
cost_gain = {'class': 1, 'bbox': 5, 'giou': 2, 'mask': 1, 'dice': 1}
@ -45,8 +48,8 @@ class HungarianMatcher(nn.Module):
def forward(self, pred_bboxes, pred_scores, gt_bboxes, gt_cls, gt_groups, masks=None, gt_mask=None):
"""
Forward pass for HungarianMatcher. This function computes costs based on prediction and ground truth
(classification cost, L1 cost between boxes and GIoU cost between boxes) and finds the optimal matching
between predictions and ground truth based on these costs.
(classification cost, L1 cost between boxes and GIoU cost between boxes) and finds the optimal matching between
predictions and ground truth based on these costs.
Args:
pred_bboxes (Tensor): Predicted bounding boxes with shape [batch_size, num_queries, 4].
@ -153,9 +156,9 @@ def get_cdn_group(batch,
box_noise_scale=1.0,
training=False):
"""
Get contrastive denoising training group. This function creates a contrastive denoising training group with
positive and negative samples from the ground truths (gt). It applies noise to the class labels and bounding
box coordinates, and returns the modified labels, bounding boxes, attention mask and meta information.
Get contrastive denoising training group. This function creates a contrastive denoising training group with positive
and negative samples from the ground truths (gt). It applies noise to the class labels and bounding box coordinates,
and returns the modified labels, bounding boxes, attention mask and meta information.
Args:
batch (dict): A dict that includes 'gt_cls' (torch.Tensor with shape [num_gts, ]), 'gt_bboxes'
@ -191,12 +194,12 @@ def get_cdn_group(batch,
gt_bbox = batch['bboxes'] # bs*num, 4
b_idx = batch['batch_idx']
# each group has positive and negative queries.
# Each group has positive and negative queries.
dn_cls = gt_cls.repeat(2 * num_group) # (2*num_group*bs*num, )
dn_bbox = gt_bbox.repeat(2 * num_group, 1) # 2*num_group*bs*num, 4
dn_b_idx = b_idx.repeat(2 * num_group).view(-1) # (2*num_group*bs*num, )
# positive and negative mask
# Positive and negative mask
# (bs*num*num_group, ), the second total_num*num_group part as negative samples
neg_idx = torch.arange(total_num * num_group, dtype=torch.long, device=gt_bbox.device) + num_group * total_num
@ -220,10 +223,9 @@ def get_cdn_group(batch,
known_bbox += rand_part * diff
known_bbox.clip_(min=0.0, max=1.0)
dn_bbox = xyxy2xywh(known_bbox)
dn_bbox = inverse_sigmoid(dn_bbox)
dn_bbox = torch.logit(dn_bbox, eps=1e-6) # inverse sigmoid
# total denoising queries
num_dn = int(max_nums * 2 * num_group)
num_dn = int(max_nums * 2 * num_group) # total denoising queries
# class_embed = torch.cat([class_embed, torch.zeros([1, class_embed.shape[-1]], device=class_embed.device)])
dn_cls_embed = class_embed[dn_cls] # bs*num * 2 * num_group, 256
padding_cls = torch.zeros(bs, num_dn, dn_cls_embed.shape[-1], device=gt_cls.device)
@ -256,9 +258,3 @@ def get_cdn_group(batch,
return padding_cls.to(class_embed.device), padding_bbox.to(class_embed.device), attn_mask.to(
class_embed.device), dn_meta
def inverse_sigmoid(x, eps=1e-6):
"""Inverse sigmoid function."""
x = x.clip(min=0., max=1.)
return torch.log(x / (1 - x + eps) + eps)