def predict(image):
image = image_loader(image=image)
BACKBONE = IR_50(INPUT_SIZE)
HEAD = ArcFace(in_features=EMBEDDING_SIZE,
out_features=1000,
device_id=GPU_ID)
BACKBONE = BACKBONE.to(DEVICE)
HEAD = HEAD.to(DEVICE)
BACKBONE.load_state_dict(
torch.load('./trained_model/Backbone_IR_50_ArcFace_30.pth'))
HEAD.load_state_dict(
torch.load('./trained_model/Head_IR_50_ArcFace_30.pth'))
BACKBONE.eval()
HEAD.eval()
image = image.to(DEVICE)
bs, ncrops, c, h, w = image.size()
inputs = image.view(-1, c, h, w)
features = BACKBONE(inputs)
outputs = HEAD(features, None)
outputs = outputs.view(bs, ncrops, -1).mean(1)
top_probs, top_labs = outputs.data.topk(1)
top_labs = top_labs.cpu().numpy()
top_probs = top_probs.cpu().numpy()
return int(top_labs), float(top_probs)
backbone_paras_only_bn, backbone_paras_wo_bn = separate_resnet_bn_paras(BACKBONE) # separate batch_norm parameters from others; do not do weight decay for batch_norm parameters to improve the generalizability
_, head_paras_wo_bn = separate_resnet_bn_paras(HEAD)
OPTIMIZER = optim.SGD([{'params': backbone_paras_wo_bn + head_paras_wo_bn, 'weight_decay': WEIGHT_DECAY}, {'params': backbone_paras_only_bn}], lr = LR, momentum = MOMENTUM)
print("=" * 60)
print(OPTIMIZER)
print("Optimizer Generated")
print("=" * 60)
# optionally resume from a checkpoint
if BACKBONE_RESUME_ROOT and HEAD_RESUME_ROOT:
print("=" * 60)
if os.path.isfile(BACKBONE_RESUME_ROOT) and os.path.isfile(HEAD_RESUME_ROOT):
print("Loading Backbone Checkpoint '{}'".format(BACKBONE_RESUME_ROOT))
BACKBONE.load_state_dict(torch.load(BACKBONE_RESUME_ROOT))
print("Loading Head Checkpoint '{}'".format(HEAD_RESUME_ROOT))
HEAD.load_state_dict(torch.load(HEAD_RESUME_ROOT))
else:
print("No Checkpoint Found at '{}' and '{}'. Please Have a Check or Continue to Train from Scratch".format(BACKBONE_RESUME_ROOT, HEAD_RESUME_ROOT))
print("=" * 60)
if MULTI_GPU:
# multi-GPU setting
BACKBONE = nn.DataParallel(BACKBONE, device_ids = GPU_ID)
BACKBONE = BACKBONE.to(DEVICE)
HEAD = nn.DataParallel(HEAD, device_ids = GPU_ID)
HEAD = HEAD.to(DEVICE)
else:
# single-GPU setting
BACKBONE = BACKBONE.to(DEVICE)
HEAD = HEAD.to(DEVICE)
backbone_paras_only_bn, backbone_paras_wo_bn = separate_resnet_bn_paras(BACKBONE) # separate batch_norm parameters from others; do not do weight decay for batch_norm parameters to improve the generalizability
_, head_paras_wo_bn = separate_resnet_bn_paras(HEAD)
OPTIMIZER = optim.SGD([{'params': backbone_paras_wo_bn + head_paras_wo_bn, 'weight_decay': WEIGHT_DECAY}, {'params': backbone_paras_only_bn}], lr = LR, momentum = MOMENTUM)
print("=" * 60)
print(OPTIMIZER)
print("Optimizer Generated")
print("=" * 60)
# optionally resume from a checkpoint
if BACKBONE_RESUME_ROOT and HEAD_RESUME_ROOT:
print("=" * 60)
if os.path.isfile(BACKBONE_RESUME_ROOT) and os.path.isfile(HEAD_RESUME_ROOT):
print("Loading Backbone Checkpoint '{}'".format(BACKBONE_RESUME_ROOT))
BACKBONE.load_state_dict(torch.load(BACKBONE_RESUME_ROOT, map_location=lambda storage, loc: storage.cuda()))
print("Loading Head Checkpoint '{}'".format(HEAD_RESUME_ROOT))
HEAD.load_state_dict(torch.load(HEAD_RESUME_ROOT, map_location=lambda storage, loc: storage.cuda()))
else:
print("No Checkpoint Found at '{}' and '{}'. Please Have a Check or Continue to Train from Scratch".format(BACKBONE_RESUME_ROOT, HEAD_RESUME_ROOT))
print("=" * 60)
# multi-GPU setting
BACKBONE = BACKBONE.cuda()
BACKBONE = nn.parallel.DistributedDataParallel(BACKBONE, device_ids = [args.local_rank], output_device=args.local_rank, find_unused_parameters=True)
HEAD = HEAD.cuda()
HEAD = nn.parallel.DistributedDataParallel(HEAD, device_ids = [args.local_rank], output_device=args.local_rank, find_unused_parameters=True)
# BACKBONE = nn.parallel.DistributedDataParallel(BACKBONE)
# LOSS = nn.parallel.DistributedDataParallel(LOSS, device_ids = [args.local_rank], output_device=args.local_rank)
# HEAD = nn.parallel.DistributedDataParallel(HEAD)