This commit is contained in:
Glenn Jocher 2018-09-22 21:50:01 +02:00
parent b93839dea7
commit bd3f617129
1 changed files with 16 additions and 16 deletions

View File

@ -158,24 +158,24 @@ class YOLOLayer(nn.Module):
tx, ty, tw, th, mask, tcls = tx.cuda(), ty.cuda(), tw.cuda(), th.cuda(), mask.cuda(), tcls.cuda() tx, ty, tw, th, mask, tcls = tx.cuda(), ty.cuda(), tw.cuda(), th.cuda(), mask.cuda(), tcls.cuda()
# Mask outputs to ignore non-existing objects (but keep confidence predictions) # Mask outputs to ignore non-existing objects (but keep confidence predictions)
nM = mask.sum().float() nT = sum([len(x) for x in targets]) # number of targets
batch_size = len(targets) nM = mask.sum().float() # number of anchors (assigned to targets)
nT = sum([len(x) for x in targets]) nB = len(targets) # batch size
if nM > 0: if nM > 0:
lx = 5 * MSELoss(x[mask], tx[mask]) lx = (5 / nB) * MSELoss(x[mask], tx[mask])
ly = 5 * MSELoss(y[mask], ty[mask]) ly = (5 / nB) * MSELoss(y[mask], ty[mask])
lw = 5 * MSELoss(w[mask], tw[mask]) lw = (5 / nB) * MSELoss(w[mask], tw[mask])
lh = 5 * MSELoss(h[mask], th[mask]) lh = (5 / nB) * MSELoss(h[mask], th[mask])
lconf = BCEWithLogitsLoss1(pred_conf[mask], mask[mask].float()) lconf = (1 / nB) * BCEWithLogitsLoss1(pred_conf[mask], mask[mask].float())
lcls = nM * CrossEntropyLoss(pred_cls[mask], torch.argmax(tcls, 1)) lcls = (1 * nM / nB) * CrossEntropyLoss(pred_cls[mask], torch.argmax(tcls, 1))
# lcls = nM * BCEWithLogitsLoss2(pred_cls[mask], tcls.float()) # lcls = (1 * nM / nB) * BCEWithLogitsLoss2(pred_cls[mask], tcls.float())
else: else:
lx, ly, lw, lh, lcls, lconf = FT([0]), FT([0]), FT([0]), FT([0]), FT([0]), FT([0]) lx, ly, lw, lh, lcls, lconf = FT([0]), FT([0]), FT([0]), FT([0]), FT([0]), FT([0])
lconf += 0.5 * nM * BCEWithLogitsLoss2(pred_conf[~mask], mask[~mask].float()) lconf += (0.5 * nM / nB) * BCEWithLogitsLoss2(pred_conf[~mask], mask[~mask].float())
loss = (lx + ly + lw + lh + lconf + lcls) / batch_size loss = lx + ly + lw + lh + lconf + lcls
# Sum False Positives from unnasigned anchors # Sum False Positives from unnasigned anchors
i = torch.sigmoid(pred_conf[~mask]) > 0.99 i = torch.sigmoid(pred_conf[~mask]) > 0.99