((h // 64) * 64, (w // 64) * 64))
x = power2resize(x)
z = largeforward(net, x.unsqueeze(0))
z = globalresize(z)
z = (z[0, 1, :, :] > z[0, 0, :, :]).float()
cm[k] += miniworld.confusion(y, z, size=size)
if False:
nextI = len(os.listdir("build"))
debug = miniworld.torchTOpil(globalresize(x))
debug = PIL.Image.fromarray(numpy.uint8(debug))
debug.save("build/" + str(nextI) + "_x.png")
debug = y.float()
debug = debug * 2 * (1 - miniworld.isborder(y, size=size))
debug = debug + miniworld.isborder(y, size=size)
debug *= 127
debug = debug.cpu().numpy()
debug = PIL.Image.fromarray(numpy.uint8(debug))
debug.save("build/" + str(nextI) + "_y.png")
debug = z.cpu().numpy() * 255
debug = PIL.Image.fromarray(numpy.uint8(debug))
debug.save("build/" + str(nextI) + "_z.png")
print("perf=", miniworld.perf(cm[k]))
perfs = miniworld.perf(cm)
print("miniworld", perfs[-1])
print(perfs)
numpy.savetxt(name,
inter0, inter1 = (y0 * z0 * D).sum(), (y1 * z1 * D).sum()
union0, union1 = ((y0 + z1 * y0) * D).sum(), ((y1 + z0 * y1) * D).sum()
iou0, iou1 = (inter0 + eps) / (union0 + eps), (inter1 + eps) / (union1 + eps)
iou = 0.5 * (iou0 + iou1)
return 1 - iou
for i in range(nbbatchs):
x, y, batchchoise, _ = dataset.getBatch(batchsize)
x, y, batchchoise = x.cuda(), y.cuda(), batchchoise.cuda()
z = net(x)
if flagborder:
D = 1 - miniworld.isborder(y, size=size)
else:
if size == 0:
D = 1 + 9 * miniworld.isborder(y, size=1)
else:
border = miniworld.isborder(y, size=1)
borderbis = miniworld.isborder(y, size=2)
borderbis = (borderbis == 1).float() * (border == 0).float()
D = 1 - border + 9 * borderbis
CE = criterion(z, y)
CE = torch.mean(CE * D)
dice = diceloss(y, z, D)
loss = CE + dice
with torch.no_grad():
inter0, inter1 = (y0 * z0 * D).sum(), (y1 * z1 * D).sum()
union0, union1 = ((y0 + z1 * y0) * D).sum(), ((y1 + z0 * y1) * D).sum()
iou0, iou1 = (inter0 + eps) / (union0 + eps), (inter1 + eps) / (union1 +
eps)
iou = 0.5 * (iou0 + iou1)
return 1 - iou
for i in range(nbbatchs):
x, y, batchchoise, _ = dataset.getBatch(batchsize)
x, y, batchchoise = x.cuda(), y.cuda(), batchchoise.cuda()
z = net(x)
D = 1 - miniworld.isborder(y)
CE = criterion(z, y)
CE = torch.mean(CE * D)
dice = diceloss(y, z, D)
loss = CE + dice
with torch.no_grad():
printloss += loss.clone().detach()
z = (z[:, 1, :, :] > z[:, 0, :, :]).clone().detach().float()
for j in range(batchsize):
stats[batchchoise[j]] += miniworld.confusion(y[j], z[j])
if i < 10:
print(i, "/", nbbatchs, printloss)
if i < 1000 and i % 100 == 99:
print(i, "/", nbbatchs, printloss / 100)
((h // 64) * 64, (w // 64) * 64))
x = power2resize(x)
z = largeforward(net, x.unsqueeze(0))
z = globalresize(z)
z = (z[0, 1, :, :] > z[0, 0, :, :]).float()
cm[k] += miniworld.confusion(y, z)
if False:
nextI = len(os.listdir("build"))
debug = miniworld.torchTOpil(globalresize(x))
debug = PIL.Image.fromarray(numpy.uint8(debug))
debug.save("build/" + str(nextI) + "_x.png")
debug = y.float()
debug = debug * 2 * (1 - miniworld.isborder(y))
debug = debug + miniworld.isborder(y)
debug *= 127
debug = debug.cpu().numpy()
debug = PIL.Image.fromarray(numpy.uint8(debug))
debug.save("build/" + str(nextI) + "_y.png")
debug = z.cpu().numpy() * 255
debug = PIL.Image.fromarray(numpy.uint8(debug))
debug.save("build/" + str(nextI) + "_z.png")
print("perf=", miniworld.perf(cm[k]))
numpy.savetxt("build/tmp.txt", miniworld.perf(cm).cpu().numpy())
perfs = miniworld.perf(cm)
print(perfs)