Module narya.models.torch_models

Classes

class ReIdModel (pretrained=False, input_shape=(256, 128), class_num=751)

Class for Torch Models responsible for the player reIdentification. It takes the image of a player and predict an embedding.

Arguments

pretrained: Boolean, if the model is loaded pretrained on ImageNet or not input_shape: Tuple, shape of the model's input class_num: Integer, number of class to predict (number of outputs in the last layer) Call arguments: input_img: a np.array of shape input_shape

Expand source code

class ReIdModel:
    """Class for Torch Models responsible for the player reIdentification. It takes the image of a player and predict an
    embedding.

    Arguments:
        pretrained: Boolean, if the model is loaded pretrained on ImageNet or not
        input_shape: Tuple, shape of the model's input 
        class_num: Integer, number of class to predict (number of outputs in the last layer)
    Call arguments:
        input_img: a np.array of shape input_shape
    """

    def __init__(self, pretrained=False, input_shape=(256, 128), class_num=751):

        self.pretrained = pretrained
        self.model = ft_net(class_num=class_num, pretrained=pretrained)
        self.input_shape = input_shape
        self.preprocessing = _build_reid_preprocessing(input_shape)
        self.class_num = class_num

    def __call__(self, input_img):

        img = self.preprocessing(input_img)

        embedding = self.model(img)
        return to_numpy(embedding)

    def load_weights(self, weights_path):
        try:
            self.model.load_state_dict(torch.load(weights_path))
            self.model.eval()
            print("Succesfully loaded weights from {}".format(weights_path))
        except:
            orig_weights = "from Imagenet" if self.pretrained else "Randomly"
            self.model.eval()
            print(
                "Could not load weights from {}, weights will be loaded {}".format(
                    weights_path, orig_weights
                )
            )

    def _get_embeddings(self, image, bbox, score, cid, score_tresh, input_shape):
        nb_samples = bbox.shape[1]
        id_feature = np.ones((1, nb_samples, self.class_num))
        resized_image = cv2.resize(image, (512, 512))
        for i in range(nb_samples):
            x_1 = int(bbox[0][i][0])
            y_1 = int(bbox[0][i][1])
            x_2 = int(bbox[0][i][2])
            y_2 = int(bbox[0][i][3])
            player_img = resized_image[y_1:y_2, x_1:x_2]
            valid_img = player_img.size > 0
            if score[0][i][0] > 0.4 and valid_img:
                embedding = self(player_img)[0]
                id_feature[0][i] = embedding
            else:
                id_feature[0][i] = np.zeros((self.class_num,))
        return id_feature

Methods

def load_weights(self, weights_path)

Expand source code

def load_weights(self, weights_path):
    try:
        self.model.load_state_dict(torch.load(weights_path))
        self.model.eval()
        print("Succesfully loaded weights from {}".format(weights_path))
    except:
        orig_weights = "from Imagenet" if self.pretrained else "Randomly"
        self.model.eval()
        print(
            "Could not load weights from {}, weights will be loaded {}".format(
                weights_path, orig_weights
            )
        )

class ft_net (class_num, droprate=0.5, stride=2, pretrained=False)

Class for custom resnet50 torch model. It is the backbone used in the Re Identification model.

Arguments

class_num: Integer, number of class to predict (number of outputs in the last layer) droprate: float in [0,1], the Dropout Rate stride: Integer, stride parameter in the model pretrained: Boolean, if the model is loaded pretrained on ImageNet or not

Initializes internal Module state, shared by both nn.Module and ScriptModule.

Expand source code

class ft_net(nn.Module):
    """Class for custom resnet50 torch model. It is the backbone used in the Re Identification model.

    Arguments:
        class_num: Integer, number of class to predict (number of outputs in the last layer)
        droprate: float in [0,1], the Dropout Rate
        stride: Integer, stride parameter in the model
        pretrained: Boolean, if the model is loaded pretrained on ImageNet or not

    """

    def __init__(self, class_num, droprate=0.5, stride=2, pretrained=False):

        super(ft_net, self).__init__()
        model_ft = models.resnet50(pretrained=pretrained)

        self.pretrained = pretrained

        if stride == 1:
            self.model.layer4[0].downsample[0].stride = (1, 1)
            self.model.layer4[0].conv2.stride = (1, 1)
        model_ft.avgpool = nn.AdaptiveAvgPool2d((1, 1))
        self.model = model_ft
        self.classifier = ClassBlock(2048, class_num, droprate)

    def forward(self, x):
        x = self.model.conv1(x)
        x = self.model.bn1(x)
        x = self.model.relu(x)
        x = self.model.maxpool(x)
        x = self.model.layer1(x)
        x = self.model.layer2(x)
        x = self.model.layer3(x)
        x = self.model.layer4(x)
        x = self.model.avgpool(x)
        x = x.view(x.size(0), x.size(1))
        x = self.classifier(x)
        return x

Ancestors

• torch.nn.modules.module.Module

Methods

def forward(self, x)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the :class:Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

Expand source code

def forward(self, x):
    x = self.model.conv1(x)
    x = self.model.bn1(x)
    x = self.model.relu(x)
    x = self.model.maxpool(x)
    x = self.model.layer1(x)
    x = self.model.layer2(x)
    x = self.model.layer3(x)
    x = self.model.layer4(x)
    x = self.model.avgpool(x)
    x = x.view(x.size(0), x.size(1))
    x = self.classifier(x)
    return x