def evaluate_F_score(val_loader, net, thresh, Fpc, Fc):
for vi, data in enumerate(val_loader):
input, mask, img_names = data
assert len(input.size()) == 4 and len(mask.size()) == 3
assert input.size()[2:] == mask.size()[1:]
input, mask_cuda = input.cuda(), mask.cuda()
with torch.no_grad():
seg_out = net(input)
seg_predictions = seg_out.data.max(1)[1].cpu()
print('evaluating: %d / %d' % (vi + 1, len(val_loader)))
_Fpc, _Fc = eval_mask_boundary(seg_predictions.numpy(), mask.numpy(), args.dataset_cls.num_classes,
bound_th=float(thresh))
Fc += _Fc
Fpc += _Fpc
del seg_out, vi, data
if args.apex:
Fc_tensor = torch.cuda.FloatTensor(Fc)
torch.distributed.all_reduce(Fc_tensor, op=torch.distributed.ReduceOp.SUM)
Fc = Fc_tensor.cpu().numpy()
Fpc_tensor = torch.cuda.FloatTensor(Fpc)
torch.distributed.all_reduce(Fpc_tensor, op=torch.distributed.ReduceOp.SUM)
Fpc = Fpc_tensor.cpu().numpy()
if args.local_rank == 0:
logging.info('Threshold: ' + thresh)
logging.info('F_Score: ' + str(np.sum(Fpc / Fc) / args.dataset_cls.num_classes))
logging.info('F_Score (Classwise): ' + str(Fpc / Fc))
return Fpc
def evaluate(val_loader, net):
'''
Runs the evaluation loop and prints F score
val_loader: Data loader for validation
net: thet network
return:
'''
net.eval()
for thresh in args.eval_thresholds.split(','):
mf_score1 = AverageMeter()
mf_pc_score1 = AverageMeter()
ap_score1 = AverageMeter()
ap_pc_score1 = AverageMeter()
Fpc = np.zeros((args.dataset_cls.num_classes))
Fc = np.zeros((args.dataset_cls.num_classes))
for vi, data in enumerate(val_loader):
input, mask, edge, img_names = data
assert len(input.size()) == 4 and len(mask.size()) == 3
assert input.size()[2:] == mask.size()[1:]
h, w = mask.size()[1:]
batch_pixel_size = input.size(0) * input.size(2) * input.size(3)
input, mask_cuda, edge_cuda = input.cuda(), mask.cuda(), edge.cuda(
)
with torch.no_grad():
seg_out, edge_out = net(input)
seg_predictions = seg_out.data.max(1)[1].cpu()
edge_predictions = edge_out.max(1)[0].cpu()
logging.info('evaluating: %d / %d' % (vi + 1, len(val_loader)))
_Fpc, _Fc = eval_mask_boundary(seg_predictions.numpy(),
mask.numpy(),
args.dataset_cls.num_classes,
bound_th=float(thresh))
Fc += _Fc
Fpc += _Fpc
del seg_out, edge_out, vi, data
logging.info('Threshold: ' + thresh)
logging.info('F_Score: ' +
str(np.sum(Fpc / Fc) / args.dataset_cls.num_classes))
logging.info('F_Score (Classwise): ' + str(Fpc / Fc))