Unverified 提交 ca9babb8 authored 作者: Glenn Jocher's avatar Glenn Jocher 提交者: GitHub

Add ComputeLoss() class (#1950)

上级 f4a78e1b
...@@ -13,7 +13,6 @@ from models.experimental import attempt_load ...@@ -13,7 +13,6 @@ from models.experimental import attempt_load
from utils.datasets import create_dataloader from utils.datasets import create_dataloader
from utils.general import coco80_to_coco91_class, check_dataset, check_file, check_img_size, check_requirements, \ from utils.general import coco80_to_coco91_class, check_dataset, check_file, check_img_size, check_requirements, \
box_iou, non_max_suppression, scale_coords, xyxy2xywh, xywh2xyxy, set_logging, increment_path, colorstr box_iou, non_max_suppression, scale_coords, xyxy2xywh, xywh2xyxy, set_logging, increment_path, colorstr
from utils.loss import compute_loss
from utils.metrics import ap_per_class, ConfusionMatrix from utils.metrics import ap_per_class, ConfusionMatrix
from utils.plots import plot_images, output_to_target, plot_study_txt from utils.plots import plot_images, output_to_target, plot_study_txt
from utils.torch_utils import select_device, time_synchronized from utils.torch_utils import select_device, time_synchronized
...@@ -36,7 +35,8 @@ def test(data, ...@@ -36,7 +35,8 @@ def test(data,
save_hybrid=False, # for hybrid auto-labelling save_hybrid=False, # for hybrid auto-labelling
save_conf=False, # save auto-label confidences save_conf=False, # save auto-label confidences
plots=True, plots=True,
log_imgs=0): # number of logged images log_imgs=0, # number of logged images
compute_loss=None):
# Initialize/load model and set device # Initialize/load model and set device
training = model is not None training = model is not None
...@@ -111,8 +111,8 @@ def test(data, ...@@ -111,8 +111,8 @@ def test(data,
t0 += time_synchronized() - t t0 += time_synchronized() - t
# Compute loss # Compute loss
if training: if compute_loss:
loss += compute_loss([x.float() for x in train_out], targets, model)[1][:3] # box, obj, cls loss += compute_loss([x.float() for x in train_out], targets)[1][:3] # box, obj, cls
# Run NMS # Run NMS
targets[:, 2:] *= torch.Tensor([width, height, width, height]).to(device) # to pixels targets[:, 2:] *= torch.Tensor([width, height, width, height]).to(device) # to pixels
......
...@@ -29,7 +29,7 @@ from utils.general import labels_to_class_weights, increment_path, labels_to_ima ...@@ -29,7 +29,7 @@ from utils.general import labels_to_class_weights, increment_path, labels_to_ima
fitness, strip_optimizer, get_latest_run, check_dataset, check_file, check_git_status, check_img_size, \ fitness, strip_optimizer, get_latest_run, check_dataset, check_file, check_git_status, check_img_size, \
check_requirements, print_mutation, set_logging, one_cycle, colorstr check_requirements, print_mutation, set_logging, one_cycle, colorstr
from utils.google_utils import attempt_download from utils.google_utils import attempt_download
from utils.loss import compute_loss from utils.loss import ComputeLoss
from utils.plots import plot_images, plot_labels, plot_results, plot_evolution from utils.plots import plot_images, plot_labels, plot_results, plot_evolution
from utils.torch_utils import ModelEMA, select_device, intersect_dicts, torch_distributed_zero_first from utils.torch_utils import ModelEMA, select_device, intersect_dicts, torch_distributed_zero_first
...@@ -227,6 +227,7 @@ def train(hyp, opt, device, tb_writer=None, wandb=None): ...@@ -227,6 +227,7 @@ def train(hyp, opt, device, tb_writer=None, wandb=None):
results = (0, 0, 0, 0, 0, 0, 0) # P, R, mAP@.5, mAP@.5-.95, val_loss(box, obj, cls) results = (0, 0, 0, 0, 0, 0, 0) # P, R, mAP@.5, mAP@.5-.95, val_loss(box, obj, cls)
scheduler.last_epoch = start_epoch - 1 # do not move scheduler.last_epoch = start_epoch - 1 # do not move
scaler = amp.GradScaler(enabled=cuda) scaler = amp.GradScaler(enabled=cuda)
compute_loss = ComputeLoss(model) # init loss class
logger.info(f'Image sizes {imgsz} train, {imgsz_test} test\n' logger.info(f'Image sizes {imgsz} train, {imgsz_test} test\n'
f'Using {dataloader.num_workers} dataloader workers\n' f'Using {dataloader.num_workers} dataloader workers\n'
f'Logging results to {save_dir}\n' f'Logging results to {save_dir}\n'
...@@ -286,7 +287,7 @@ def train(hyp, opt, device, tb_writer=None, wandb=None): ...@@ -286,7 +287,7 @@ def train(hyp, opt, device, tb_writer=None, wandb=None):
# Forward # Forward
with amp.autocast(enabled=cuda): with amp.autocast(enabled=cuda):
pred = model(imgs) # forward pred = model(imgs) # forward
loss, loss_items = compute_loss(pred, targets.to(device), model) # loss scaled by batch_size loss, loss_items = compute_loss(pred, targets.to(device)) # loss scaled by batch_size
if rank != -1: if rank != -1:
loss *= opt.world_size # gradient averaged between devices in DDP mode loss *= opt.world_size # gradient averaged between devices in DDP mode
if opt.quad: if opt.quad:
...@@ -344,7 +345,8 @@ def train(hyp, opt, device, tb_writer=None, wandb=None): ...@@ -344,7 +345,8 @@ def train(hyp, opt, device, tb_writer=None, wandb=None):
dataloader=testloader, dataloader=testloader,
save_dir=save_dir, save_dir=save_dir,
plots=plots and final_epoch, plots=plots and final_epoch,
log_imgs=opt.log_imgs if wandb else 0) log_imgs=opt.log_imgs if wandb else 0,
compute_loss=compute_loss)
# Write # Write
with open(results_file, 'a') as f: with open(results_file, 'a') as f:
......
...@@ -85,119 +85,133 @@ class QFocalLoss(nn.Module): ...@@ -85,119 +85,133 @@ class QFocalLoss(nn.Module):
return loss return loss
def compute_loss(p, targets, model): # predictions, targets, model class ComputeLoss:
device = targets.device # Compute losses
lcls, lbox, lobj = torch.zeros(1, device=device), torch.zeros(1, device=device), torch.zeros(1, device=device) def __init__(self, model, autobalance=False):
tcls, tbox, indices, anchors = build_targets(p, targets, model) # targets super(ComputeLoss, self).__init__()
h = model.hyp # hyperparameters device = next(model.parameters()).device # get model device
h = model.hyp # hyperparameters
# Define criteria
BCEcls = nn.BCEWithLogitsLoss(pos_weight=torch.tensor([h['cls_pw']], device=device)) # weight=model.class_weights) # Define criteria
BCEobj = nn.BCEWithLogitsLoss(pos_weight=torch.tensor([h['obj_pw']], device=device)) BCEcls = nn.BCEWithLogitsLoss(pos_weight=torch.tensor([h['cls_pw']], device=device))
BCEobj = nn.BCEWithLogitsLoss(pos_weight=torch.tensor([h['obj_pw']], device=device))
# Class label smoothing https://arxiv.org/pdf/1902.04103.pdf eqn 3
cp, cn = smooth_BCE(eps=0.0) # Class label smoothing https://arxiv.org/pdf/1902.04103.pdf eqn 3
self.cp, self.cn = smooth_BCE(eps=0.0)
# Focal loss
g = h['fl_gamma'] # focal loss gamma # Focal loss
if g > 0: g = h['fl_gamma'] # focal loss gamma
BCEcls, BCEobj = FocalLoss(BCEcls, g), FocalLoss(BCEobj, g) if g > 0:
BCEcls, BCEobj = FocalLoss(BCEcls, g), FocalLoss(BCEobj, g)
# Losses
nt = 0 # number of targets det = model.module.model[-1] if is_parallel(model) else model.model[-1] # Detect() module
balance = [4.0, 1.0, 0.3, 0.1, 0.03] # P3-P7 self.balance = {3: [3.67, 1.0, 0.43], 4: [3.78, 1.0, 0.39, 0.22], 5: [3.88, 1.0, 0.37, 0.17, 0.10]}[det.nl]
for i, pi in enumerate(p): # layer index, layer predictions # self.balance = [1.0] * det.nl
b, a, gj, gi = indices[i] # image, anchor, gridy, gridx self.ssi = (det.stride == 16).nonzero(as_tuple=False).item() # stride 16 index
tobj = torch.zeros_like(pi[..., 0], device=device) # target obj self.BCEcls, self.BCEobj, self.gr, self.hyp, self.autobalance = BCEcls, BCEobj, model.gr, h, autobalance
for k in 'na', 'nc', 'nl', 'anchors':
n = b.shape[0] # number of targets setattr(self, k, getattr(det, k))
if n:
nt += n # cumulative targets def __call__(self, p, targets): # predictions, targets, model
ps = pi[b, a, gj, gi] # prediction subset corresponding to targets device = targets.device
lcls, lbox, lobj = torch.zeros(1, device=device), torch.zeros(1, device=device), torch.zeros(1, device=device)
# Regression tcls, tbox, indices, anchors = self.build_targets(p, targets) # targets
pxy = ps[:, :2].sigmoid() * 2. - 0.5
pwh = (ps[:, 2:4].sigmoid() * 2) ** 2 * anchors[i] # Losses
pbox = torch.cat((pxy, pwh), 1) # predicted box for i, pi in enumerate(p): # layer index, layer predictions
iou = bbox_iou(pbox.T, tbox[i], x1y1x2y2=False, CIoU=True) # iou(prediction, target) b, a, gj, gi = indices[i] # image, anchor, gridy, gridx
lbox += (1.0 - iou).mean() # iou loss tobj = torch.zeros_like(pi[..., 0], device=device) # target obj
# Objectness n = b.shape[0] # number of targets
tobj[b, a, gj, gi] = (1.0 - model.gr) + model.gr * iou.detach().clamp(0).type(tobj.dtype) # iou ratio if n:
ps = pi[b, a, gj, gi] # prediction subset corresponding to targets
# Classification
if model.nc > 1: # cls loss (only if multiple classes) # Regression
t = torch.full_like(ps[:, 5:], cn, device=device) # targets pxy = ps[:, :2].sigmoid() * 2. - 0.5
t[range(n), tcls[i]] = cp pwh = (ps[:, 2:4].sigmoid() * 2) ** 2 * anchors[i]
lcls += BCEcls(ps[:, 5:], t) # BCE pbox = torch.cat((pxy, pwh), 1) # predicted box
iou = bbox_iou(pbox.T, tbox[i], x1y1x2y2=False, CIoU=True) # iou(prediction, target)
# Append targets to text file lbox += (1.0 - iou).mean() # iou loss
# with open('targets.txt', 'a') as file:
# [file.write('%11.5g ' * 4 % tuple(x) + '\n') for x in torch.cat((txy[i], twh[i]), 1)] # Objectness
tobj[b, a, gj, gi] = (1.0 - self.gr) + self.gr * iou.detach().clamp(0).type(tobj.dtype) # iou ratio
lobj += BCEobj(pi[..., 4], tobj) * balance[i] # obj loss
# Classification
lbox *= h['box'] if self.nc > 1: # cls loss (only if multiple classes)
lobj *= h['obj'] t = torch.full_like(ps[:, 5:], self.cn, device=device) # targets
lcls *= h['cls'] t[range(n), tcls[i]] = self.cp
bs = tobj.shape[0] # batch size lcls += self.BCEcls(ps[:, 5:], t) # BCE
loss = lbox + lobj + lcls # Append targets to text file
return loss * bs, torch.cat((lbox, lobj, lcls, loss)).detach() # with open('targets.txt', 'a') as file:
# [file.write('%11.5g ' * 4 % tuple(x) + '\n') for x in torch.cat((txy[i], twh[i]), 1)]
def build_targets(p, targets, model): obji = self.BCEobj(pi[..., 4], tobj)
# Build targets for compute_loss(), input targets(image,class,x,y,w,h) lobj += obji * self.balance[i] # obj loss
det = model.module.model[-1] if is_parallel(model) else model.model[-1] # Detect() module if self.autobalance:
na, nt = det.na, targets.shape[0] # number of anchors, targets self.balance[i] = self.balance[i] * 0.9999 + 0.0001 / obji.detach().item()
tcls, tbox, indices, anch = [], [], [], []
gain = torch.ones(7, device=targets.device) # normalized to gridspace gain if self.autobalance:
ai = torch.arange(na, device=targets.device).float().view(na, 1).repeat(1, nt) # same as .repeat_interleave(nt) self.balance = [x / self.balance[self.ssi] for x in self.balance]
targets = torch.cat((targets.repeat(na, 1, 1), ai[:, :, None]), 2) # append anchor indices lbox *= self.hyp['box']
lobj *= self.hyp['obj']
g = 0.5 # bias lcls *= self.hyp['cls']
off = torch.tensor([[0, 0], bs = tobj.shape[0] # batch size
[1, 0], [0, 1], [-1, 0], [0, -1], # j,k,l,m
# [1, 1], [1, -1], [-1, 1], [-1, -1], # jk,jm,lk,lm loss = lbox + lobj + lcls
], device=targets.device).float() * g # offsets return loss * bs, torch.cat((lbox, lobj, lcls, loss)).detach()
for i in range(det.nl): def build_targets(self, p, targets):
anchors = det.anchors[i] # Build targets for compute_loss(), input targets(image,class,x,y,w,h)
gain[2:6] = torch.tensor(p[i].shape)[[3, 2, 3, 2]] # xyxy gain na, nt = self.na, targets.shape[0] # number of anchors, targets
tcls, tbox, indices, anch = [], [], [], []
# Match targets to anchors gain = torch.ones(7, device=targets.device) # normalized to gridspace gain
t = targets * gain ai = torch.arange(na, device=targets.device).float().view(na, 1).repeat(1, nt) # same as .repeat_interleave(nt)
if nt: targets = torch.cat((targets.repeat(na, 1, 1), ai[:, :, None]), 2) # append anchor indices
# Matches
r = t[:, :, 4:6] / anchors[:, None] # wh ratio g = 0.5 # bias
j = torch.max(r, 1. / r).max(2)[0] < model.hyp['anchor_t'] # compare off = torch.tensor([[0, 0],
# j = wh_iou(anchors, t[:, 4:6]) > model.hyp['iou_t'] # iou(3,n)=wh_iou(anchors(3,2), gwh(n,2)) [1, 0], [0, 1], [-1, 0], [0, -1], # j,k,l,m
t = t[j] # filter # [1, 1], [1, -1], [-1, 1], [-1, -1], # jk,jm,lk,lm
], device=targets.device).float() * g # offsets
# Offsets
for i in range(self.nl):
anchors = self.anchors[i]
gain[2:6] = torch.tensor(p[i].shape)[[3, 2, 3, 2]] # xyxy gain
# Match targets to anchors
t = targets * gain
if nt:
# Matches
r = t[:, :, 4:6] / anchors[:, None] # wh ratio
j = torch.max(r, 1. / r).max(2)[0] < self.hyp['anchor_t'] # compare
# j = wh_iou(anchors, t[:, 4:6]) > model.hyp['iou_t'] # iou(3,n)=wh_iou(anchors(3,2), gwh(n,2))
t = t[j] # filter
# Offsets
gxy = t[:, 2:4] # grid xy
gxi = gain[[2, 3]] - gxy # inverse
j, k = ((gxy % 1. < g) & (gxy > 1.)).T
l, m = ((gxi % 1. < g) & (gxi > 1.)).T
j = torch.stack((torch.ones_like(j), j, k, l, m))
t = t.repeat((5, 1, 1))[j]
offsets = (torch.zeros_like(gxy)[None] + off[:, None])[j]
else:
t = targets[0]
offsets = 0
# Define
b, c = t[:, :2].long().T # image, class
gxy = t[:, 2:4] # grid xy gxy = t[:, 2:4] # grid xy
gxi = gain[[2, 3]] - gxy # inverse gwh = t[:, 4:6] # grid wh
j, k = ((gxy % 1. < g) & (gxy > 1.)).T gij = (gxy - offsets).long()
l, m = ((gxi % 1. < g) & (gxi > 1.)).T gi, gj = gij.T # grid xy indices
j = torch.stack((torch.ones_like(j), j, k, l, m))
t = t.repeat((5, 1, 1))[j] # Append
offsets = (torch.zeros_like(gxy)[None] + off[:, None])[j] a = t[:, 6].long() # anchor indices
else: indices.append((b, a, gj.clamp_(0, gain[3] - 1), gi.clamp_(0, gain[2] - 1))) # image, anchor, grid indices
t = targets[0] tbox.append(torch.cat((gxy - gij, gwh), 1)) # box
offsets = 0 anch.append(anchors[a]) # anchors
tcls.append(c) # class
# Define
b, c = t[:, :2].long().T # image, class return tcls, tbox, indices, anch
gxy = t[:, 2:4] # grid xy
gwh = t[:, 4:6] # grid wh
gij = (gxy - offsets).long()
gi, gj = gij.T # grid xy indices
# Append
a = t[:, 6].long() # anchor indices
indices.append((b, a, gj.clamp_(0, gain[3] - 1), gi.clamp_(0, gain[2] - 1))) # image, anchor, grid indices
tbox.append(torch.cat((gxy - gij, gwh), 1)) # box
anch.append(anchors[a]) # anchors
tcls.append(c) # class
return tcls, tbox, indices, anch
Markdown 格式
0%
您添加了 0 到此讨论。请谨慎行事。
请先完成此评论的编辑!
注册 或者 后发表评论