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

ultralytics 8.0.197 save P, R, F1 curves to metrics (#5354)

Browse source 
Signed-off-by: Glenn Jocher <glenn.jocher@ultralytics.com>
Co-authored-by: erminkev1 <83356055+erminkev1@users.noreply.github.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Andy <39454881+yermandy@users.noreply.github.com>
This commit is contained in:
Glenn Jocher 2023-10-13 02:49:31 +02:00 • committed by GitHub
parent 7fd5dcbd86
commit 12e3eef844
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
33 changed files with 337 additions and 195 deletions
Download patch file Download diff file Expand all files Collapse all files

2
ultralytics/models/fastsam/prompt.py
View file

@ -120,7 +120,7 @@ class FastSAMPrompt:
result_name = os.path.basename(ann.path)
image = ann.orig_img[..., ::-1] # BGR to RGB
original_h, original_w = ann.orig_shape
# for macOS only
# For macOS only
# plt.switch_backend('TkAgg')
plt.figure(figsize=(original_w / 100, original_h / 100))
# Add subplot with no margin.

10
ultralytics/models/fastsam/utils.py
View file

@ -42,23 +42,23 @@ def bbox_iou(box1, boxes, iou_thres=0.9, image_shape=(640, 640), raw_output=Fals
high_iou_indices (torch.Tensor): Indices of boxes with IoU > thres
"""
boxes = adjust_bboxes_to_image_border(boxes, image_shape)
# obtain coordinates for intersections
# Obtain coordinates for intersections
x1 = torch.max(box1[0], boxes[:, 0])
y1 = torch.max(box1[1], boxes[:, 1])
x2 = torch.min(box1[2], boxes[:, 2])
y2 = torch.min(box1[3], boxes[:, 3])
# compute the area of intersection
# Compute the area of intersection
intersection = (x2 - x1).clamp(0) * (y2 - y1).clamp(0)
# compute the area of both individual boxes
# Compute the area of both individual boxes
box1_area = (box1[2] - box1[0]) * (box1[3] - box1[1])
box2_area = (boxes[:, 2] - boxes[:, 0]) * (boxes[:, 3] - boxes[:, 1])
# compute the area of union
# Compute the area of union
union = box1_area + box2_area - intersection
# compute the IoU
# Compute the IoU
iou = intersection / union # Should be shape (n, )
if raw_output:
return 0 if iou.numel() == 0 else iou

4
ultralytics/models/rtdetr/val.py
View file

@ -99,10 +99,10 @@ class RTDETRValidator(DetectionValidator):
for i, bbox in enumerate(bboxes): # (300, 4)
bbox = ops.xywh2xyxy(bbox)
score, cls = scores[i].max(-1) # (300, )
# Do not need threshold for evaluation as only got 300 boxes here.
# Do not need threshold for evaluation as only got 300 boxes here
# idx = score > self.args.conf
pred = torch.cat([bbox, score[..., None], cls[..., None]], dim=-1) # filter
# sort by confidence to correctly get internal metrics.
# Sort by confidence to correctly get internal metrics
pred = pred[score.argsort(descending=True)]
outputs[i] = pred # [idx]

6
ultralytics/models/sam/modules/encoders.py
View file

@ -304,11 +304,11 @@ class PositionEmbeddingRandom(nn.Module):
def _pe_encoding(self, coords: torch.Tensor) -> torch.Tensor:
"""Positionally encode points that are normalized to [0,1]."""
# assuming coords are in [0, 1]^2 square and have d_1 x ... x d_n x 2 shape
# Assuming coords are in [0, 1]^2 square and have d_1 x ... x d_n x 2 shape
coords = 2 * coords - 1
coords = coords @ self.positional_encoding_gaussian_matrix
coords = 2 * np.pi * coords
# outputs d_1 x ... x d_n x C shape
# Outputs d_1 x ... x d_n x C shape
return torch.cat([torch.sin(coords), torch.cos(coords)], dim=-1)
def forward(self, size: Tuple[int, int]) -> torch.Tensor:
@ -429,7 +429,7 @@ class Attention(nn.Module):
self.use_rel_pos = use_rel_pos
if self.use_rel_pos:
assert (input_size is not None), 'Input size must be provided if using relative positional encoding.'
# initialize relative positional embeddings
# Initialize relative positional embeddings
self.rel_pos_h = nn.Parameter(torch.zeros(2 * input_size[0] - 1, head_dim))
self.rel_pos_w = nn.Parameter(torch.zeros(2 * input_size[1] - 1, head_dim))

20
ultralytics/models/sam/modules/tiny_encoder.py
View file

@ -172,7 +172,7 @@ class ConvLayer(nn.Module):
self.depth = depth
self.use_checkpoint = use_checkpoint
# build blocks
# Build blocks
self.blocks = nn.ModuleList([
MBConv(
dim,
@ -182,7 +182,7 @@ class ConvLayer(nn.Module):
drop_path[i] if isinstance(drop_path, list) else drop_path,
) for i in range(depth)])
# patch merging layer
# Patch merging layer
self.downsample = None if downsample is None else downsample(
input_resolution, dim=dim, out_dim=out_dim, activation=activation)
@ -393,11 +393,11 @@ class TinyViTBlock(nn.Module):
pH, pW = H + pad_b, W + pad_r
nH = pH // self.window_size
nW = pW // self.window_size
# window partition
# Window partition
x = x.view(B, nH, self.window_size, nW, self.window_size,
C).transpose(2, 3).reshape(B * nH * nW, self.window_size * self.window_size, C)
x = self.attn(x)
# window reverse
# Window reverse
x = x.view(B, nH, nW, self.window_size, self.window_size, C).transpose(2, 3).reshape(B, pH, pW, C)
if padding:
@ -467,7 +467,7 @@ class BasicLayer(nn.Module):
self.depth = depth
self.use_checkpoint = use_checkpoint
# build blocks
# Build blocks
self.blocks = nn.ModuleList([
TinyViTBlock(
dim=dim,
@ -481,7 +481,7 @@ class BasicLayer(nn.Module):
activation=activation,
) for i in range(depth)])
# patch merging layer
# Patch merging layer
self.downsample = None if downsample is None else downsample(
input_resolution, dim=dim, out_dim=out_dim, activation=activation)
@ -593,10 +593,10 @@ class TinyViT(nn.Module):
patches_resolution = self.patch_embed.patches_resolution
self.patches_resolution = patches_resolution
# stochastic depth
# Stochastic depth
dpr = [x.item() for x in torch.linspace(0, drop_path_rate, sum(depths))] # stochastic depth decay rule
# build layers
# Build layers
self.layers = nn.ModuleList()
for i_layer in range(self.num_layers):
kwargs = dict(
@ -628,7 +628,7 @@ class TinyViT(nn.Module):
self.norm_head = nn.LayerNorm(embed_dims[-1])
self.head = nn.Linear(embed_dims[-1], num_classes) if num_classes > 0 else torch.nn.Identity()
# init weights
# Init weights
self.apply(self._init_weights)
self.set_layer_lr_decay(layer_lr_decay)
self.neck = nn.Sequential(
@ -653,7 +653,7 @@ class TinyViT(nn.Module):
"""Sets the learning rate decay for each layer in the TinyViT model."""
decay_rate = layer_lr_decay
# layers -> blocks (depth)
# Layers -> blocks (depth)
depth = sum(self.depths)
lr_scales = [decay_rate ** (depth - i - 1) for i in range(depth)]

3
ultralytics/models/sam/predict.py
View file

@ -414,8 +414,7 @@ class Predictor(BasePredictor):
unchanged = unchanged and not changed
new_masks.append(torch.as_tensor(mask).unsqueeze(0))
# Give score=0 to changed masks and score=1 to unchanged masks
# so NMS will prefer ones that didn't need postprocessing
# Give score=0 to changed masks and 1 to unchanged masks so NMS prefers masks not needing postprocessing
scores.append(float(unchanged))
# Recalculate boxes and remove any new duplicates

4
ultralytics/models/utils/loss.py
View file

@ -66,7 +66,7 @@ class DETRLoss(nn.Module):
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]]
# Logits: [b, query, num_classes], gt_class: list[[n, 1]]
name_class = f'loss_class{postfix}'
bs, nq = pred_scores.shape[:2]
# one_hot = F.one_hot(targets, self.nc + 1)[..., :-1] # (bs, num_queries, num_classes)
@ -90,7 +90,7 @@ class DETRLoss(nn.Module):
"""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]]
# Boxes: [b, query, 4], gt_bbox: list[[n, 4]]
name_bbox = f'loss_bbox{postfix}'
name_giou = f'loss_giou{postfix}'

10
ultralytics/models/utils/ops.py
View file

@ -188,7 +188,7 @@ def get_cdn_group(batch,
num_group = num_dn // max_nums
num_group = 1 if num_group == 0 else num_group
# pad gt to max_num of a batch
# Pad gt to max_num of a batch
bs = len(gt_groups)
gt_cls = batch['cls'] # (bs*num, )
gt_bbox = batch['bboxes'] # bs*num, 4
@ -204,10 +204,10 @@ def get_cdn_group(batch,
neg_idx = torch.arange(total_num * num_group, dtype=torch.long, device=gt_bbox.device) + num_group * total_num
if cls_noise_ratio > 0:
# half of bbox prob
# Half of bbox prob
mask = torch.rand(dn_cls.shape) < (cls_noise_ratio * 0.5)
idx = torch.nonzero(mask).squeeze(-1)
# randomly put a new one here
# Randomly put a new one here
new_label = torch.randint_like(idx, 0, num_classes, dtype=dn_cls.dtype, device=dn_cls.device)
dn_cls[idx] = new_label
@ -240,9 +240,9 @@ def get_cdn_group(batch,
tgt_size = num_dn + num_queries
attn_mask = torch.zeros([tgt_size, tgt_size], dtype=torch.bool)
# match query cannot see the reconstruct
# Match query cannot see the reconstruct
attn_mask[num_dn:, :num_dn] = True
# reconstruct cannot see each other
# Reconstruct cannot see each other
for i in range(num_group):
if i == 0:
attn_mask[max_nums * 2 * i:max_nums * 2 * (i + 1), max_nums * 2 * (i + 1):num_dn] = True