Actions
Forum Thread Details
  • Replies 1 reply
  • Subscribers 45 subscribers
  • Views 1424 views
  • Users 0 members are here
Related

Can anyone help me solve this error I am getting

manojroy123 

import numpy as np
import tensorflow as tf
from tensorflow.keras.models import Model
from tensorflow.keras.layers import Input, Conv2D, BatchNormalization, LeakyReLU, concatenate

def tiny_yolo_v2(input_shape, num_classes):
    inputs = Input(shape=input_shape)

    # Tiny YOLOv2 architecture
    x = Conv2D(16, (3, 3), padding='same')(inputs)
    x = BatchNormalization()(x)
    x = LeakyReLU(alpha=0.1)(x)

    x = Conv2D(32, (3, 3), strides=(2, 2), padding='same')(x)
    x = BatchNormalization()(x)
    x = LeakyReLU(alpha=0.1)(x)

    x = Conv2D(64, (3, 3), strides=(2, 2), padding='same')(x)
    x = BatchNormalization()(x)
    x = LeakyReLU(alpha=0.1)(x)

    x = Conv2D(128, (3, 3), strides=(2, 2), padding='same')(x)
    x = BatchNormalization()(x)
    x = LeakyReLU(alpha=0.1)(x)

    x = Conv2D(256, (3, 3), strides=(2, 2), padding='same')(x)
    x = BatchNormalization()(x)
    x = LeakyReLU(alpha=0.1)(x)

    # Final output layer
    x = Conv2D(num_classes, (1, 1), padding='same')(x)

    model = Model(inputs, x)
    return model

def load_darknet_weights(model, weights_file):
    with open(weights_file, "rb") as f:
        header = np.fromfile(f, dtype=np.int32, count=5)  # Read header
        weights = np.fromfile(f, dtype=np.float32)  # Read weights

    layer_index = 0
    for layer in model.layers:
        if isinstance(layer, Conv2D):
            filters = layer.filters
            kernel_size = layer.kernel_size[0]
            input_dim = layer.input_shape[-1]
            
            # Load weights
            if layer.use_bias:
                start = layer_index * (filters * (kernel_size ** 2) * input_dim + filters)
            else:
                start = layer_index * (filters * (kernel_size ** 2) * input_dim)

            end = start + (filters * (kernel_size ** 2) * input_dim)
            kernel = weights[start:end].reshape((kernel_size, kernel_size, input_dim, filters))
            layer.set_weights([kernel])
            
            
            if layer.use_bias:
                start = end
                end = start + filters
                bias = weights[start:end]
                layer.set_weights([kernel, bias])

            layer_index += 1

# Specify parameters
input_shape = (416, 416, 3)
num_classes = 20  # Adjust based on your dataset

# Create the Tiny YOLOv2 model
model = tiny_yolo_v2(input_shape, num_classes)

# Load weights from Darknet
weights_file = 'yolov2-tiny.weights'  # Path to your weights file
load_darknet_weights(model, weights_file)

# Save the model in H5 format
model.save('tiny_yolo_v2.h5')

print("Model saved to 'tiny_yolo_v2.h5'")

I am getting this error

> %Run 'file reading.py'
2024-10-10 14:42:55.306715: I tensorflow/core/util/port.cc:153] oneDNN custom operations are on. You may see slightly different numerical results due to floating-point round-off errors from different computation orders. To turn them off, set the environment variable `TF_ENABLE_ONEDNN_OPTS=0`.
2024-10-10 14:43:07.741864: I tensorflow/core/util/port.cc:153] oneDNN custom operations are on. You may see slightly different numerical results due to floating-point round-off errors from different computation orders. To turn them off, set the environment variable `TF_ENABLE_ONEDNN_OPTS=0`.
C:\Users\manojroy\AppData\Roaming\Python\Python310\site-packages\keras\src\layers\activations\leaky_relu.py:41: UserWarning: Argument `alpha` is deprecated. Use `negative_slope` instead.
warnings.warn(
Traceback (most recent call last):
File "C:\Users\manojroy\Desktop\CircuitDigest\AssistiveTechForBlind\Yolo2Tiny\file reading.py", line 76, in <module>
load_darknet_weights(model, weights_file)
File "C:\Users\manojroy\Desktop\CircuitDigest\AssistiveTechForBlind\Yolo2Tiny\file reading.py", line 46, in load_darknet_weights
input_dim = layer.input_shape[-1]
AttributeError: 'Conv2D' object has no attribute 'input_shape'

the code is for convertion of yolov2 tiny weights to tensorflow *.h5 weights. If some one has any better solutions please let me know about it.

Parents

  • Edit your code like this and try:

    import numpy as np
import tensorflow as tf
from tensorflow.keras.models import Model
from tensorflow.keras.layers import Input, Conv2D, BatchNormalization, LeakyReLU

def tiny_yolo_v2(input_shape, num_classes):
    inputs = Input(shape=input_shape)

    # Tiny YOLOv2 architecture
    x = Conv2D(16, (3, 3), padding='same', use_bias=False)(inputs)
    x = BatchNormalization()(x)
    x = LeakyReLU(negative_slope=0.1)(x)

    x = Conv2D(32, (3, 3), strides=(2, 2), padding='same', use_bias=False)(x)
    x = BatchNormalization()(x)
    x = LeakyReLU(negative_slope=0.1)(x)

    x = Conv2D(64, (3, 3), strides=(2, 2), padding='same', use_bias=False)(x)
    x = BatchNormalization()(x)
    x = LeakyReLU(negative_slope=0.1)(x)

    x = Conv2D(128, (3, 3), strides=(2, 2), padding='same', use_bias=False)(x)
    x = BatchNormalization()(x)
    x = LeakyReLU(negative_slope=0.1)(x)

    x = Conv2D(256, (3, 3), strides=(2, 2), padding='same', use_bias=False)(x)
    x = BatchNormalization()(x)
    x = LeakyReLU(negative_slope=0.1)(x)

    # Final output layer
    x = Conv2D(num_classes, (1, 1), padding='same', use_bias=True)(x)

    model = Model(inputs, x)
    return model

def load_darknet_weights(model, weights_file):
    with open(weights_file, "rb") as f:
        header = np.fromfile(f, dtype=np.int32, count=5)  # Read header
        weights = np.fromfile(f, dtype=np.float32)  # Read weights

    ptr = 0  # Pointer to track the current position in weights array

    for layer in model.layers:
        if isinstance(layer, Conv2D):
            # Get the number of filters, kernel size, and input dimensions
            filters = layer.filters
            kernel_size = layer.kernel_size[0]
            input_dim = layer.input.shape[-1]

            # Calculate the size of the kernel weights
            kernel_weights_size = filters * input_dim * kernel_size * kernel_size
            # Extract kernel weights and reshape
            kernel_weights = weights[ptr:ptr + kernel_weights_size]
            kernel_weights = kernel_weights.reshape((kernel_size, kernel_size, input_dim, filters))
            ptr += kernel_weights_size

            if layer.use_bias:
                # If the layer has bias, load them
                bias_weights = weights[ptr:ptr + filters]
                ptr += filters
                layer.set_weights([kernel_weights, bias_weights])
            else:
                # If the layer doesn't use bias, it likely has BatchNorm
                # Load BatchNorm parameters: [gamma, beta, mean, variance]
                bn_layer = model.layers[model.layers.index(layer) + 1]
                if isinstance(bn_layer, BatchNormalization):
                    bn_weights = weights[ptr:ptr + 4 * filters]
                    gamma = bn_weights[0:filters]
                    beta = bn_weights[filters:2 * filters]
                    mean = bn_weights[2 * filters:3 * filters]
                    variance = bn_weights[3 * filters:4 * filters]
                    bn_layer.set_weights([gamma, beta, mean, variance])
                    ptr += 4 * filters

# Specify parameters
input_shape = (416, 416, 3)
num_classes = 20  # Adjust based on your dataset

# Create the Tiny YOLOv2 model
model = tiny_yolo_v2(input_shape, num_classes)

# Load weights from Darknet
weights_file = 'yolov2-tiny.weights'  # Path to your weights file
load_darknet_weights(model, weights_file)

# Save the model in H5 format
model.save('tiny_yolo_v2.h5')

print("Model saved to 'tiny_yolo_v2.h5'")

Reply

  • Edit your code like this and try:

    import numpy as np
import tensorflow as tf
from tensorflow.keras.models import Model
from tensorflow.keras.layers import Input, Conv2D, BatchNormalization, LeakyReLU

def tiny_yolo_v2(input_shape, num_classes):
    inputs = Input(shape=input_shape)

    # Tiny YOLOv2 architecture
    x = Conv2D(16, (3, 3), padding='same', use_bias=False)(inputs)
    x = BatchNormalization()(x)
    x = LeakyReLU(negative_slope=0.1)(x)

    x = Conv2D(32, (3, 3), strides=(2, 2), padding='same', use_bias=False)(x)
    x = BatchNormalization()(x)
    x = LeakyReLU(negative_slope=0.1)(x)

    x = Conv2D(64, (3, 3), strides=(2, 2), padding='same', use_bias=False)(x)
    x = BatchNormalization()(x)
    x = LeakyReLU(negative_slope=0.1)(x)

    x = Conv2D(128, (3, 3), strides=(2, 2), padding='same', use_bias=False)(x)
    x = BatchNormalization()(x)
    x = LeakyReLU(negative_slope=0.1)(x)

    x = Conv2D(256, (3, 3), strides=(2, 2), padding='same', use_bias=False)(x)
    x = BatchNormalization()(x)
    x = LeakyReLU(negative_slope=0.1)(x)

    # Final output layer
    x = Conv2D(num_classes, (1, 1), padding='same', use_bias=True)(x)

    model = Model(inputs, x)
    return model

def load_darknet_weights(model, weights_file):
    with open(weights_file, "rb") as f:
        header = np.fromfile(f, dtype=np.int32, count=5)  # Read header
        weights = np.fromfile(f, dtype=np.float32)  # Read weights

    ptr = 0  # Pointer to track the current position in weights array

    for layer in model.layers:
        if isinstance(layer, Conv2D):
            # Get the number of filters, kernel size, and input dimensions
            filters = layer.filters
            kernel_size = layer.kernel_size[0]
            input_dim = layer.input.shape[-1]

            # Calculate the size of the kernel weights
            kernel_weights_size = filters * input_dim * kernel_size * kernel_size
            # Extract kernel weights and reshape
            kernel_weights = weights[ptr:ptr + kernel_weights_size]
            kernel_weights = kernel_weights.reshape((kernel_size, kernel_size, input_dim, filters))
            ptr += kernel_weights_size

            if layer.use_bias:
                # If the layer has bias, load them
                bias_weights = weights[ptr:ptr + filters]
                ptr += filters
                layer.set_weights([kernel_weights, bias_weights])
            else:
                # If the layer doesn't use bias, it likely has BatchNorm
                # Load BatchNorm parameters: [gamma, beta, mean, variance]
                bn_layer = model.layers[model.layers.index(layer) + 1]
                if isinstance(bn_layer, BatchNormalization):
                    bn_weights = weights[ptr:ptr + 4 * filters]
                    gamma = bn_weights[0:filters]
                    beta = bn_weights[filters:2 * filters]
                    mean = bn_weights[2 * filters:3 * filters]
                    variance = bn_weights[3 * filters:4 * filters]
                    bn_layer.set_weights([gamma, beta, mean, variance])
                    ptr += 4 * filters

# Specify parameters
input_shape = (416, 416, 3)
num_classes = 20  # Adjust based on your dataset

# Create the Tiny YOLOv2 model
model = tiny_yolo_v2(input_shape, num_classes)

# Load weights from Darknet
weights_file = 'yolov2-tiny.weights'  # Path to your weights file
load_darknet_weights(model, weights_file)

# Save the model in H5 format
model.save('tiny_yolo_v2.h5')

print("Model saved to 'tiny_yolo_v2.h5'")

Children
No Data