import torch import torch.nn as nn import torch.nn.functional as F from ..general import xywh2xyxy from ..loss import FocalLoss, smooth_BCE from ..metrics import bbox_iou from ..torch_utils import de_parallel from .general import crop_mask class ComputeLoss: # Compute losses def __init__(self, model, autobalance=False, overlap=False): self.sort_obj_iou = False self.overlap = overlap 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)) BCEobj = nn.BCEWithLogitsLoss(pos_weight=torch.tensor([h['obj_pw']], device=device)) # Class label smoothing https://arxiv.org/pdf/1902.04103.pdf eqn 3 self.cp, self.cn = smooth_BCE(eps=h.get('label_smoothing', 0.0)) # positive, negative BCE targets # Focal loss g = h['fl_gamma'] # focal loss gamma if g > 0: BCEcls, BCEobj = FocalLoss(BCEcls, g), FocalLoss(BCEobj, g) m = de_parallel(model).model[-1] # Detect() module self.balance = {3: [4.0, 1.0, 0.4]}.get(m.nl, [4.0, 1.0, 0.25, 0.06, 0.02]) # P3-P7 self.ssi = list(m.stride).index(16) if autobalance else 0 # stride 16 index self.BCEcls, self.BCEobj, self.gr, self.hyp, self.autobalance = BCEcls, BCEobj, 1.0, h, autobalance self.na = m.na # number of anchors self.nc = m.nc # number of classes self.nl = m.nl # number of layers self.nm = m.nm # number of masks self.anchors = m.anchors self.device = device def __call__(self, preds, targets, masks): # predictions, targets, model p, proto = preds bs, nm, mask_h, mask_w = proto.shape # batch size, number of masks, mask height, mask width lcls = torch.zeros(1, device=self.device) lbox = torch.zeros(1, device=self.device) lobj = torch.zeros(1, device=self.device) lseg = torch.zeros(1, device=self.device) tcls, tbox, indices, anchors, tidxs, xywhn = self.build_targets(p, targets) # targets # Losses for i, pi in enumerate(p): # layer index, layer predictions b, a, gj, gi = indices[i] # image, anchor, gridy, gridx tobj = torch.zeros(pi.shape[:4], dtype=pi.dtype, device=self.device) # target obj n = b.shape[0] # number of targets if n: pxy, pwh, _, pcls, pmask = pi[b, a, gj, gi].split((2, 2, 1, self.nc, nm), 1) # subset of predictions # Box regression pxy = pxy.sigmoid() * 2 - 0.5 pwh = (pwh.sigmoid() * 2) ** 2 * anchors[i] pbox = torch.cat((pxy, pwh), 1) # predicted box iou = bbox_iou(pbox, tbox[i], CIoU=True).squeeze() # iou(prediction, target) lbox += (1.0 - iou).mean() # iou loss # Objectness iou = iou.detach().clamp(0).type(tobj.dtype) if self.sort_obj_iou: j = iou.argsort() b, a, gj, gi, iou = b[j], a[j], gj[j], gi[j], iou[j] if self.gr < 1: iou = (1.0 - self.gr) + self.gr * iou tobj[b, a, gj, gi] = iou # iou ratio # Classification if self.nc > 1: # cls loss (only if multiple classes) t = torch.full_like(pcls, self.cn, device=self.device) # targets t[range(n), tcls[i]] = self.cp lcls += self.BCEcls(pcls, t) # BCE # Mask regression if tuple(masks.shape[-2:]) != (mask_h, mask_w): # downsample masks = F.interpolate(masks[None], (mask_h, mask_w), mode='nearest')[0] marea = xywhn[i][:, 2:].prod(1) # mask width, height normalized mxyxy = xywh2xyxy(xywhn[i] * torch.tensor([mask_w, mask_h, mask_w, mask_h], device=self.device)) for bi in b.unique(): j = b == bi # matching index if self.overlap: mask_gti = torch.where(masks[bi][None] == tidxs[i][j].view(-1, 1, 1), 1.0, 0.0) else: mask_gti = masks[tidxs[i]][j] lseg += self.single_mask_loss(mask_gti, pmask[j], proto[bi], mxyxy[j], marea[j]) obji = self.BCEobj(pi[..., 4], tobj) lobj += obji * self.balance[i] # obj loss if self.autobalance: self.balance[i] = self.balance[i] * 0.9999 + 0.0001 / obji.detach().item() if self.autobalance: self.balance = [x / self.balance[self.ssi] for x in self.balance] lbox *= self.hyp['box'] lobj *= self.hyp['obj'] lcls *= self.hyp['cls'] lseg *= self.hyp['box'] / bs loss = lbox + lobj + lcls + lseg return loss * bs, torch.cat((lbox, lseg, lobj, lcls)).detach() def single_mask_loss(self, gt_mask, pred, proto, xyxy, area): # Mask loss for one image pred_mask = (pred @ proto.view(self.nm, -1)).view(-1, *proto.shape[1:]) # (n,32) @ (32,80,80) -> (n,80,80) loss = F.binary_cross_entropy_with_logits(pred_mask, gt_mask, reduction='none') return (crop_mask(loss, xyxy).mean(dim=(1, 2)) / area).mean() def build_targets(self, p, targets): # Build targets for compute_loss(), input targets(image,class,x,y,w,h) na, nt = self.na, targets.shape[0] # number of anchors, targets tcls, tbox, indices, anch, tidxs, xywhn = [], [], [], [], [], [] gain = torch.ones(8, device=self.device) # normalized to gridspace gain ai = torch.arange(na, device=self.device).float().view(na, 1).repeat(1, nt) # same as .repeat_interleave(nt) if self.overlap: batch = p[0].shape[0] ti = [] for i in range(batch): num = (targets[:, 0] == i).sum() # find number of targets of each image ti.append(torch.arange(num, device=self.device).float().view(1, num).repeat(na, 1) + 1) # (na, num) ti = torch.cat(ti, 1) # (na, nt) else: ti = torch.arange(nt, device=self.device).float().view(1, nt).repeat(na, 1) targets = torch.cat((targets.repeat(na, 1, 1), ai[..., None], ti[..., None]), 2) # append anchor indices g = 0.5 # bias off = torch.tensor( [ [0, 0], [1, 0], [0, 1], [-1, 0], [0, -1], # j,k,l,m # [1, 1], [1, -1], [-1, 1], [-1, -1], # jk,jm,lk,lm ], device=self.device).float() * g # offsets for i in range(self.nl): anchors, shape = self.anchors[i], p[i].shape gain[2:6] = torch.tensor(shape)[[3, 2, 3, 2]] # xyxy gain # Match targets to anchors t = targets * gain # shape(3,n,7) 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 bc, gxy, gwh, at = t.chunk(4, 1) # (image, class), grid xy, grid wh, anchors (a, tidx), (b, c) = at.long().T, bc.long().T # anchors, image, class gij = (gxy - offsets).long() gi, gj = gij.T # grid indices # Append indices.append((b, a, gj.clamp_(0, shape[2] - 1), gi.clamp_(0, shape[3] - 1))) # image, anchor, grid tbox.append(torch.cat((gxy - gij, gwh), 1)) # box anch.append(anchors[a]) # anchors tcls.append(c) # class tidxs.append(tidx) xywhn.append(torch.cat((gxy, gwh), 1) / gain[2:6]) # xywh normalized return tcls, tbox, indices, anch, tidxs, xywhn