Python thop.profile方法代碼示例

# 需要導入模塊: import thop [as 別名]
# 或者: from thop import profile [as 別名]
def main():
    args = get_args()

    model_kwargs = {}
    if args.rectify:
        model_kwargs['rectified_conv'] = True
        model_kwargs['rectify_avg'] = args.rectify_avg

    model = encoding.models.get_model(args.model, **model_kwargs)
    print(model)

    dummy_images = torch.rand(1, 3, args.crop_size, args.crop_size)

    #count_ops(model, dummy_images, verbose=False)
    macs, params = profile(model, inputs=(dummy_images, ))
    macs, params = clever_format([macs, params], "%.3f") 

    print(f"macs: {macs}, params: {params}") 

# 需要導入模塊: import thop [as 別名]
# 或者: from thop import profile [as 別名]
def __init__(self, loadweights=True, downsample=4, model_path='pretrained_model/mobilenetv2_1.0-0c6065bc.pth'):
        super(mobilenetv2_base, self).__init__()

        model = MobileNetV2(width_mult=1.0)

        if loadweights:
            model.load_state_dict(torch.load(model_path))

        #if downsample == 4:
        #    self.feature = nn.Sequential(model.features[:14])
        #elif downsample == 5:
        #    self.feature = nn.Sequential(model.features)

        self.feature3 = nn.Sequential(model.features[:7])
        self.feature4 = nn.Sequential(model.features[7:14])
        self.feature5 = nn.Sequential(model.features[14:])

        #flops, params = profile(self.feature, input_size=(1, 3, 256,256)) 

# 需要導入模塊: import thop [as 別名]
# 或者: from thop import profile [as 別名]
def print_flops(model):
    shape = None
    if config["dataset"] in ["Cifar10", "Cifar100"]:
        shape = (1, 3, 32, 32)
    else:
        print(f"Unknown dataset {config['dataset']} input size to compute # FLOPS")
        return

    try:
        from thop import profile
    except:
        print("Please `pip install thop` to compute # FLOPS")
        return

    model = model.train()
    input_data = torch.rand(*shape)
    num_flops, num_params = profile(model, inputs=(input_data, ))
    print("Number of FLOPS:", human_format(num_flops)) 

# 需要導入模塊: import thop [as 別名]
# 或者: from thop import profile [as 別名]
def show_flops_params(model, device, input_shape=[1, 3, 1024, 2048]):
    #summary(model, tuple(input_shape[1:]), device=device)
    input = torch.randn(*input_shape).to(torch.device(device))
    flops, params = profile(model, inputs=(input,), verbose=False)

    logging.info('{} flops: {:.3f}G input shape is {}, params: {:.3f}M'.format(
        model.__class__.__name__, flops / 1000000000, input_shape[1:], params / 1000000)) 

# 需要導入模塊: import thop [as 別名]
# 或者: from thop import profile [as 別名]
def calcFlops(model, input):
    flops, params = profile(model, inputs=(input, ))
    flops, params = clever_format([flops, params], "%.3f")
    print('flops: {} \nparameters: {}'.format(flops, params))
    return flops, params 

# 需要導入模塊: import thop [as 別名]
# 或者: from thop import profile [as 別名]
def _flops(h, w, C_in, C_out, kernel_size=3, stride=1, padding=None, dilation=1, groups=1, bias=False):
        layer = ConvNorm(C_in, C_out, kernel_size, stride, padding, dilation, groups, bias, slimmable=False)
        flops, params = profile(layer, inputs=(torch.randn(1, C_in, h, w),), verbose=False)
        return flops 

# 需要導入模塊: import thop [as 別名]
# 或者: from thop import profile [as 別名]
def objective(trial):

    # Generate the model.
    model = define_model(trial).to(DEVICE)

    # Generate the optimizers.
    optimizer_name = trial.suggest_categorical("optimizer", ["Adam", "RMSprop", "SGD"])
    lr = trial.suggest_uniform("lr", 1e-5, 1e-1)
    optimizer = getattr(optim, optimizer_name)(model.parameters(), lr=lr)

    # Get the MNIST dataset.
    train_loader, val_loader = get_mnist()

    # Training of the model.
    model.train()
    for epoch in range(EPOCHS):
        for batch_idx, (data, target) in enumerate(train_loader):
            data, target = data.view(-1, 28 * 28).to(DEVICE), target.to(DEVICE)

            optimizer.zero_grad()
            output = model(data)
            loss = F.nll_loss(output, target)
            loss.backward()
            optimizer.step()

    # Validation of the model.
    model.eval()
    correct = 0
    with torch.no_grad():
        for batch_idx, (data, target) in enumerate(val_loader):
            data, target = data.view(-1, 28 * 28).to(DEVICE), target.to(DEVICE)
            output = model(data)
            pred = output.argmax(dim=1, keepdim=True)  # Get the index of the max log-probability.
            correct += pred.eq(target.view_as(pred)).sum().item()

    accuracy = correct / N_VAL_EXAMPLES

    flops, _params = thop.profile(model, inputs=(torch.randn(1, 28 * 28),), verbose=False)
    return flops, accuracy 

# 需要導入模塊: import thop [as 別名]
# 或者: from thop import profile [as 別名]
def main():
    global args
    args = parser.parse_args()
    model = models.__dict__[args.arch]()    
    print(model)
    input = torch.randn(1, 3, 224, 224)
    model.train()
    # model.eval()
    flops, params = profile(model, inputs=(input, ))
    print("flops = ", flops)
    print("params = ", params)
    flops, params = clever_format([flops, params], "%.3f")
    print("flops = ", flops)
    print("params = ", params) 

# 需要導入模塊: import thop [as 別名]
# 或者: from thop import profile [as 別名]
def main():
    create_exp_dir(config.save, scripts_to_save=glob.glob('*.py')+glob.glob('*.sh'))

    log_format = '%(asctime)s %(message)s'
    logging.basicConfig(stream=sys.stdout, level=logging.INFO, format=log_format, datefmt='%m/%d %I:%M:%S %p')
    fh = logging.FileHandler(os.path.join(config.save, 'log.txt'))
    fh.setFormatter(logging.Formatter(log_format))
    logging.getLogger().addHandler(fh)
    logging.info("args = %s", str(config))
    # preparation ################
    torch.backends.cudnn.enabled = True
    torch.backends.cudnn.benchmark = True
    seed = config.seed
    np.random.seed(seed)
    torch.manual_seed(seed)
    if torch.cuda.is_available():
        torch.cuda.manual_seed(seed)

    # Model #######################################
    lasts = []
    for idx, arch_idx in enumerate(config.arch_idx):
        if config.load_epoch == "last":
            state = torch.load(os.path.join(config.load_path, "arch_%d.pt"%arch_idx))
        else:
            state = torch.load(os.path.join(config.load_path, "arch_%d_%d.pt"%(arch_idx, int(config.load_epoch))))

        model = Network(
            [state["alpha_%d_0"%arch_idx].detach(), state["alpha_%d_1"%arch_idx].detach(), state["alpha_%d_2"%arch_idx].detach()],
            [None, state["beta_%d_1"%arch_idx].detach(), state["beta_%d_2"%arch_idx].detach()],
            [state["ratio_%d_0"%arch_idx].detach(), state["ratio_%d_1"%arch_idx].detach(), state["ratio_%d_2"%arch_idx].detach()],
            num_classes=config.num_classes, layers=config.layers, Fch=config.Fch, width_mult_list=config.width_mult_list, stem_head_width=config.stem_head_width[idx], ignore_skip=arch_idx==0)

        mIoU02 = state["mIoU02"]; latency02 = state["latency02"]; obj02 = objective_acc_lat(mIoU02, latency02)
        mIoU12 = state["mIoU12"]; latency12 = state["latency12"]; obj12 = objective_acc_lat(mIoU12, latency12)
        if obj02 > obj12: last = [2, 0]
        else: last = [2, 1]
        lasts.append(last)
        model.build_structure(last)
        logging.info("net: " + str(model))
        for b in last:
            if len(config.width_mult_list) > 1:
                plot_op(getattr(model, "ops%d"%b), getattr(model, "path%d"%b), width=getattr(model, "widths%d"%b), head_width=config.stem_head_width[idx][1], F_base=config.Fch).savefig(os.path.join(config.save, "ops_%d_%d.png"%(arch_idx,b)), bbox_inches="tight")
            else:
                plot_op(getattr(model, "ops%d"%b), getattr(model, "path%d"%b), F_base=config.Fch).savefig(os.path.join(config.save, "ops_%d_%d.png"%(arch_idx,b)), bbox_inches="tight")
        plot_path_width(model.lasts, model.paths, model.widths).savefig(os.path.join(config.save, "path_width%d.png"%arch_idx))
        plot_path_width([2, 1, 0], [model.path2, model.path1, model.path0], [model.widths2, model.widths1, model.widths0]).savefig(os.path.join(config.save, "path_width_all%d.png"%arch_idx))
        flops, params = profile(model, inputs=(torch.randn(1, 3, 1024, 2048),), verbose=False)
        logging.info("params = %fMB, FLOPs = %fGB", params / 1e6, flops / 1e9)
        logging.info("ops:" + str(model.ops))
        logging.info("path:" + str(model.paths))
        model = model.cuda()
        #####################################################
        print(config.save)
        latency = compute_latency(model, (1, 3, config.image_height, config.image_width))
        logging.info("FPS:" + str(1000./latency)) 

# 需要導入模塊: import thop [as 別名]
# 或者: from thop import profile [as 別名]
def count_flops_params(model: nn.Module, input_size, verbose=True):
    """
    Count FLOPs and Params of the given model.
    This function would identify the mask on the module
    and take the pruned shape into consideration.
    Note that, for sturctured pruning, we only identify
    the remained filters according to its mask, which
    not taking the pruned input channels into consideration,
    so the calculated FLOPs will be larger than real number.

    Parameters
    ---------
    model : nn.Module
        target model.
    input_size: list, tuple
        the input shape of data


    Returns
    -------
    flops: float
        total flops of the model
    params:
        total params of the model
    """

    assert input_size is not None

    device = next(model.parameters()).device
    inputs = torch.randn(input_size).to(device)

    hook_module_list = []
    prev_m = None
    for m in model.modules():
        weight_mask = None
        m_type = type(m)
        if m_type in custom_ops:
            if isinstance(prev_m, PrunerModuleWrapper):
                weight_mask = prev_m.weight_mask

            m.register_buffer('weight_mask', weight_mask)
            hook_module_list.append(m)
        prev_m = m

    flops, params = profile(model, inputs=(inputs, ), custom_ops=custom_ops, verbose=verbose)

    for m in hook_module_list:
        m._buffers.pop("weight_mask")

    return flops, params 

# 需要導入模塊: import thop [as 別名]
# 或者: from thop import profile [as 別名]
def evaluate_detector(args):
    """Evaluate directional point detector."""
    args.cuda = not args.disable_cuda and torch.cuda.is_available()
    device = torch.device('cuda:' + str(args.gpu_id) if args.cuda else 'cpu')
    torch.set_grad_enabled(False)

    dp_detector = DirectionalPointDetector(
        3, args.depth_factor, config.NUM_FEATURE_MAP_CHANNEL).to(device)
    if args.detector_weights:
        dp_detector.load_state_dict(torch.load(args.detector_weights))
    dp_detector.eval()

    psdataset = ParkingSlotDataset(args.dataset_directory)
    logger = util.Logger(enable_visdom=args.enable_visdom)

    total_loss = 0
    position_errors = []
    direction_errors = []
    ground_truths_list = []
    predictions_list = []
    for iter_idx, (image, marking_points) in enumerate(psdataset):
        ground_truths_list.append(marking_points)

        image = torch.unsqueeze(image, 0).to(device)
        prediction = dp_detector(image)
        objective, gradient = generate_objective([marking_points], device)
        loss = (prediction - objective) ** 2
        total_loss += torch.sum(loss*gradient).item()

        pred_points = get_predicted_points(prediction[0], 0.01)
        predictions_list.append(pred_points)

        dists, angles = collect_error(marking_points, pred_points,
                                      config.CONFID_THRESH_FOR_POINT)
        position_errors += dists
        direction_errors += angles

        logger.log(iter=iter_idx, total_loss=total_loss)

    precisions, recalls = util.calc_precision_recall(
        ground_truths_list, predictions_list, match_marking_points)
    average_precision = util.calc_average_precision(precisions, recalls)
    if args.enable_visdom:
        logger.plot_curve(precisions, recalls)

    sample = torch.randn(1, 3, config.INPUT_IMAGE_SIZE,
                         config.INPUT_IMAGE_SIZE)
    flops, params = profile(dp_detector, inputs=(sample.to(device), ))
    logger.log(average_loss=total_loss / len(psdataset),
               average_precision=average_precision,
               flops=flops,
               params=params) 

# 需要導入模塊: import thop [as 別名]
# 或者: from thop import profile [as 別名]
def bulid_up_network(config,criterion):

    # if config.model.use_backbone:
    #     logger.info("backbone of architecture is {}".format(config.model.backbone_net_name))

    if config.model.backbone_net_name=="resnet":
        backbone = BackBone_ResNet(config,is_train=True)

    if config.model.backbone_net_name=="mobilenet_v2":
        backbone = BackBone_MobileNet(config,is_train=True)

    if config.model.backbone_net_name=="meta_arch":
        logger.info("backbone:{}".format(config.model.backbone))
        backbone = Backbone_Arch(criterion,**config.model.backbone)

    if config.model.backbone_net_name=="hrnet":
        backbone = BackBone_HRNet(config,is_train=True)
       
    Arch = Body_Part_Representation(config.model.keypoints_num,  criterion, backbone, **config.model.subnetwork_config)


    if config.model.use_pretrained:
        Arch.load_pretrained(config.model.pretrained)

    
        
    logger.info("\n\nbackbone: params and flops")
    logger.info(get_model_summary(backbone,torch.randn(1, 3, config.model.input_size.h,config.model.input_size.w)))

    logger.info("\n\nwhole architecture: params and flops")
    logger.info(get_model_summary(Arch,torch.randn(1, 3, config.model.input_size.h,config.model.input_size.w)))

    logger.info("=========== thop statistics ==========")
    dump = torch.randn(1, 3, config.model.input_size.h,config.model.input_size.w)
    flops, params = profile( backbone, inputs=(dump,),  )
    logger.info(">>> total params of BackBone: {:.2f}M\n>>> total FLOPS of Backbone: {:.3f} G\n".format(
                    (params / 1000000.0),(flops / 1000000000.0)))
    flops, params = profile(Arch, inputs=(dump,),  )
    logger.info(">>> total params of Whole Model: {:.2f}M\n>>> total FLOPS of Model: {:.3f} G\n".format(
                        (params / 1000000.0),(flops / 1000000000.0)))

    return Arch