YOLOv5 segmentation model support (#9052)
* [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Fix duplicate plots.py * Fix check_font() * # torch.use_deterministic_algorithms(True) * update doc detect->predict * Resolve precommit for segment/train and segment/val * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Resolve precommit for utils/segment * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Resolve precommit min_wh * Resolve precommit utils/segment/plots * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Resolve precommit utils/segment/general * Align NMS-seg closer to NMS * restore deterministic init_seeds code * remove easydict dependency * update * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * restore output_to_target mask * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * update * cleanup * Remove unused ImageFont import * Unified NMS * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * DetectMultiBackend compatibility * segment/predict.py update * update plot colors * fix bbox shifted * sort bbox by confidence * enable overlap by default * Merge detect/segment output_to_target() function * Start segmentation CI * fix plots * Update ci-testing.yml * fix training whitespace * optimize process mask functions (can we merge both?) * Update predict/detect * Update plot_images * Update plot_images_and_masks * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * fix * Add train to CI * fix precommit * fix precommit CI * fix precommit pycocotools * fix val float issues * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * fix masks float float issues * suppress errors * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * fix no-predictions plotting bug * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Add CSV Logger * fix val len(plot_masks) * speed up evaluation * fix process_mask * fix plots * update segment/utils build_targets * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * optimize utils/segment/general crop() * optimize utils/segment/general crop() 2 * minor updates * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * torch.where revert * downsample only if different shape * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * loss cleanup * loss cleanup 2 * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * loss cleanup 3 * update project names * Rename -seg yamls from _underscore to -dash * prepare for yolov5n-seg.pt * precommit space fix * add coco128-seg.yaml * update coco128-seg comments * cleanup val.py * Major val.py cleanup * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * precommit fix * precommit fix * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * optional pycocotools * remove CI pip install pycocotools (auto-installed now) * seg yaml fix * optimize mask_iou() and masks_iou() * threaded fix * Major train.py update * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Major segments/val/process_batch() update * yolov5/val updates from segment * process_batch numpy/tensor fix * opt-in to pycocotools with --save-json * threaded pycocotools ops for 2x speed increase * Avoid permute contiguous if possible * Add max_det=300 argument to both val.py and segment/val.py * fix onnx_dynamic * speed up pycocotools ops * faster process_mask(upsample=True) for predict * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * eliminate permutations for process_mask(upsample=True) * eliminate permute-contiguous in crop(), use native dimension order * cleanup comment * Add Proto() module * fix class count * fix anchor order * broadcast mask_gti in loss for speed * Cleanup seg loss * faster indexing * faster indexing fix * faster indexing fix2 * revert faster indexing * fix validation plotting * Loss cleanup and mxyxy simplification * Loss cleanup and mxyxy simplification 2 * revert validation plotting * replace missing tanh * Eliminate last permutation * delete unneeded .float() * Remove MaskIOULoss and crop(if HWC) * Final v6.3 SegmentationModel architecture updates * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Add support for TF export * remove debugger trace * add call * update * update * Merge master * Merge master * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Update dataloaders.py * Restore CI * Update dataloaders.py * Fix TF/TFLite export for segmentation model * Merge master * Cleanup predict.py mask plotting * cleanup scale_masks() * rename scale_masks to scale_image * cleanup/optimize plot_masks * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Add Annotator.masks() * Annotator.masks() fix * Update plots.py * Annotator mask optimization * Rename crop() to crop_mask() * Do not crop in predict.py * crop always * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Merge master * Add vid-stride from master PR * Update seg model outputs * Update seg model outputs * Add segmentation benchmarks * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Add segmentation benchmarks * Add segmentation benchmarks * Add segmentation benchmarks * Fix DetectMultiBackend for OpenVINO * update Annotator.masks * fix val plot * revert val plot * clean up * revert pil * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Fix CI error * fix predict log * remove upsample * update interpolate * fix validation plot logging * Annotator.masks() cleanup * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Remove segmentation_model definition * Restore 0.99999 decimals Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> Co-authored-by: Glenn Jocher <glenn.jocher@ultralytics.com> Co-authored-by: Laughing-q <1185102784@qq.com> Co-authored-by: Jiacong Fang <zldrobit@126.com>
This commit is contained in:
27 files changed
+3091
-71
No files matched your search
+44
-14
@@ -36,6 +36,7 @@ except ImportError:
|
||||
|
||||
|
||||
class Detect(nn.Module):
|
||||
# YOLOv5 Detect head for detection models
|
||||
stride = None # strides computed during build
|
||||
dynamic = False # force grid reconstruction
|
||||
export = False # export mode
|
||||
@@ -63,15 +64,16 @@ class Detect(nn.Module):
|
||||
if self.dynamic or self.grid[i].shape[2:4] != x[i].shape[2:4]:
|
||||
self.grid[i], self.anchor_grid[i] = self._make_grid(nx, ny, i)
|
||||
|
||||
y = x[i].sigmoid()
|
||||
y = x[i].clone()
|
||||
y[..., :5 + self.nc].sigmoid_()
|
||||
if self.inplace:
|
||||
y[..., 0:2] = (y[..., 0:2] * 2 + self.grid[i]) * self.stride[i] # xy
|
||||
y[..., 2:4] = (y[..., 2:4] * 2) ** 2 * self.anchor_grid[i] # wh
|
||||
else: # for YOLOv5 on AWS Inferentia https://github.com/ultralytics/yolov5/pull/2953
|
||||
xy, wh, conf = y.split((2, 2, self.nc + 1), 4) # y.tensor_split((2, 4, 5), 4) # torch 1.8.0
|
||||
xy, wh, etc = y.split((2, 2, self.no - 4), 4) # tensor_split((2, 4, 5), 4) if torch 1.8.0
|
||||
xy = (xy * 2 + self.grid[i]) * self.stride[i] # xy
|
||||
wh = (wh * 2) ** 2 * self.anchor_grid[i] # wh
|
||||
y = torch.cat((xy, wh, conf), 4)
|
||||
y = torch.cat((xy, wh, etc), 4)
|
||||
z.append(y.view(bs, -1, self.no))
|
||||
|
||||
return x if self.training else (torch.cat(z, 1),) if self.export else (torch.cat(z, 1), x)
|
||||
@@ -87,6 +89,23 @@ class Detect(nn.Module):
|
||||
return grid, anchor_grid
|
||||
|
||||
|
||||
class Segment(Detect):
|
||||
# YOLOv5 Segment head for segmentation models
|
||||
def __init__(self, nc=80, anchors=(), nm=32, npr=256, ch=(), inplace=True):
|
||||
super().__init__(nc, anchors, ch, inplace)
|
||||
self.nm = nm # number of masks
|
||||
self.npr = npr # number of protos
|
||||
self.no = 5 + nc + self.nm # number of outputs per anchor
|
||||
self.m = nn.ModuleList(nn.Conv2d(x, self.no * self.na, 1) for x in ch) # output conv
|
||||
self.proto = Proto(ch[0], self.npr, self.nm) # protos
|
||||
self.detect = Detect.forward
|
||||
|
||||
def forward(self, x):
|
||||
p = self.proto(x[0])
|
||||
x = self.detect(self, x)
|
||||
return (x, p) if self.training else (x[0], p) if self.export else (x[0], p, x[1])
|
||||
|
||||
|
||||
class BaseModel(nn.Module):
|
||||
# YOLOv5 base model
|
||||
def forward(self, x, profile=False, visualize=False):
|
||||
@@ -135,7 +154,7 @@ class BaseModel(nn.Module):
|
||||
# Apply to(), cpu(), cuda(), half() to model tensors that are not parameters or registered buffers
|
||||
self = super()._apply(fn)
|
||||
m = self.model[-1] # Detect()
|
||||
if isinstance(m, Detect):
|
||||
if isinstance(m, (Detect, Segment)):
|
||||
m.stride = fn(m.stride)
|
||||
m.grid = list(map(fn, m.grid))
|
||||
if isinstance(m.anchor_grid, list):
|
||||
@@ -169,11 +188,12 @@ class DetectionModel(BaseModel):
|
||||
|
||||
# Build strides, anchors
|
||||
m = self.model[-1] # Detect()
|
||||
if isinstance(m, Detect):
|
||||
if isinstance(m, (Detect, Segment)):
|
||||
s = 256 # 2x min stride
|
||||
m.inplace = self.inplace
|
||||
m.stride = torch.tensor([s / x.shape[-2] for x in self.forward(torch.empty(1, ch, s, s))]) # forward
|
||||
check_anchor_order(m) # must be in pixel-space (not grid-space)
|
||||
forward = lambda x: self.forward(x)[0] if isinstance(m, Segment) else self.forward(x)
|
||||
m.stride = torch.tensor([s / x.shape[-2] for x in forward(torch.zeros(1, ch, s, s))]) # forward
|
||||
check_anchor_order(m)
|
||||
m.anchors /= m.stride.view(-1, 1, 1)
|
||||
self.stride = m.stride
|
||||
self._initialize_biases() # only run once
|
||||
@@ -235,15 +255,21 @@ class DetectionModel(BaseModel):
|
||||
# cf = torch.bincount(torch.tensor(np.concatenate(dataset.labels, 0)[:, 0]).long(), minlength=nc) + 1.
|
||||
m = self.model[-1] # Detect() module
|
||||
for mi, s in zip(m.m, m.stride): # from
|
||||
b = mi.bias.view(m.na, -1).detach() # conv.bias(255) to (3,85)
|
||||
b[:, 4] += math.log(8 / (640 / s) ** 2) # obj (8 objects per 640 image)
|
||||
b[:, 5:] += math.log(0.6 / (m.nc - 0.999999)) if cf is None else torch.log(cf / cf.sum()) # cls
|
||||
b = mi.bias.view(m.na, -1) # conv.bias(255) to (3,85)
|
||||
b.data[:, 4] += math.log(8 / (640 / s) ** 2) # obj (8 objects per 640 image)
|
||||
b.data[:, 5:5 + m.nc] += math.log(0.6 / (m.nc - 0.99999)) if cf is None else torch.log(cf / cf.sum()) # cls
|
||||
mi.bias = torch.nn.Parameter(b.view(-1), requires_grad=True)
|
||||
|
||||
|
||||
Model = DetectionModel # retain YOLOv5 'Model' class for backwards compatibility
|
||||
|
||||
|
||||
class SegmentationModel(DetectionModel):
|
||||
# YOLOv5 segmentation model
|
||||
def __init__(self, cfg='yolov5s-seg.yaml', ch=3, nc=None, anchors=None):
|
||||
super().__init__(cfg, ch, nc, anchors)
|
||||
|
||||
|
||||
class ClassificationModel(BaseModel):
|
||||
# YOLOv5 classification model
|
||||
def __init__(self, cfg=None, model=None, nc=1000, cutoff=10): # yaml, model, number of classes, cutoff index
|
||||
@@ -284,24 +310,28 @@ def parse_model(d, ch): # model_dict, input_channels(3)
|
||||
args[j] = eval(a) if isinstance(a, str) else a # eval strings
|
||||
|
||||
n = n_ = max(round(n * gd), 1) if n > 1 else n # depth gain
|
||||
if m in (Conv, GhostConv, Bottleneck, GhostBottleneck, SPP, SPPF, DWConv, MixConv2d, Focus, CrossConv,
|
||||
BottleneckCSP, C3, C3TR, C3SPP, C3Ghost, nn.ConvTranspose2d, DWConvTranspose2d, C3x):
|
||||
if m in {
|
||||
Conv, GhostConv, Bottleneck, GhostBottleneck, SPP, SPPF, DWConv, MixConv2d, Focus, CrossConv,
|
||||
BottleneckCSP, C3, C3TR, C3SPP, C3Ghost, nn.ConvTranspose2d, DWConvTranspose2d, C3x}:
|
||||
c1, c2 = ch[f], args[0]
|
||||
if c2 != no: # if not output
|
||||
c2 = make_divisible(c2 * gw, 8)
|
||||
|
||||
args = [c1, c2, *args[1:]]
|
||||
if m in [BottleneckCSP, C3, C3TR, C3Ghost, C3x]:
|
||||
if m in {BottleneckCSP, C3, C3TR, C3Ghost, C3x}:
|
||||
args.insert(2, n) # number of repeats
|
||||
n = 1
|
||||
elif m is nn.BatchNorm2d:
|
||||
args = [ch[f]]
|
||||
elif m is Concat:
|
||||
c2 = sum(ch[x] for x in f)
|
||||
elif m is Detect:
|
||||
# TODO: channel, gw, gd
|
||||
elif m in {Detect, Segment}:
|
||||
args.append([ch[x] for x in f])
|
||||
if isinstance(args[1], int): # number of anchors
|
||||
args[1] = [list(range(args[1] * 2))] * len(f)
|
||||
if m is Segment:
|
||||
args[3] = make_divisible(args[3] * gw, 8)
|
||||
elif m is Contract:
|
||||
c2 = ch[f] * args[0] ** 2
|
||||
elif m is Expand:
|
||||
|
||||
Reference in new issue
Block a user