forked from david8862/keras-YOLOv3-model-set
-
Notifications
You must be signed in to change notification settings - Fork 0
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
- Loading branch information
Showing
11 changed files
with
241 additions
and
57 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,16 @@ | ||
| set(CMAKE_SYSTEM_NAME Linux) | ||
| set(CMAKE_SYSTEM_PROCESSOR aarch64) | ||
| set(CMAKE_LIBRARY_ARCHITECTURE aarch64-linux-gnu) | ||
|
|
||
| set(PROJECT_ROOT_DIR "${CMAKE_CURRENT_LIST_DIR}/../..") | ||
|
|
||
| set(CMAKE_C_COMPILER aarch64-linux-gnu-gcc) | ||
| set(CMAKE_CXX_COMPILER aarch64-linux-gnu-g++) | ||
|
|
||
| set(CMAKE_SYSROOT ${PROJECT_ROOT_DIR}/toolchain/aarch64-linux-gnu/${CMAKE_LIBRARY_ARCHITECTURE}/sysroot) | ||
| set(CMAKE_FIND_ROOT_PATH ${PROJECT_ROOT_DIR}/toolchain/aarch64-linux-gnu/${CMAKE_LIBRARY_ARCHITECTURE}/sysroot) | ||
|
|
||
| set(CMAKE_CROSSCOMPILING true) | ||
| set(CMAKE_FIND_ROOT_PATH_MODE_PROGRAM NEVER) | ||
| set(CMAKE_FIND_ROOT_PATH_MODE_LIBRARY ONLY) | ||
| set(CMAKE_FIND_ROOT_PATH_MODE_INCLUDE ONLY) |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,165 @@ | ||
| #!/usr/bin/env python3 | ||
| # -*- coding: utf-8 -*- | ||
| ''' | ||
| generate activation pattern image for convolution kernel of specified layers | ||
| with trained YOLO model | ||
| Reference: | ||
| https://blog.csdn.net/qq_37781464/article/details/122946523 | ||
| ''' | ||
| import os, sys, argparse | ||
| import numpy as np | ||
|
|
||
| from tensorflow.keras.models import Model, load_model | ||
| import matplotlib.pyplot as plt | ||
| import tensorflow as tf | ||
| from tensorflow.keras.layers import Conv2D, DepthwiseConv2D, SeparableConv2D | ||
| import tensorflow.keras.backend as K | ||
|
|
||
| # compatible with TF 2.x | ||
| if tf.__version__.startswith('2'): | ||
| import tensorflow.compat.v1 as tf | ||
| from tensorflow.compat.v1.keras import backend as K | ||
| tf.disable_eager_execution() | ||
|
|
||
| sys.path.append(os.path.join(os.path.dirname(os.path.realpath(__file__)), '..', '..')) | ||
| from common.utils import get_custom_objects, optimize_tf_gpu | ||
|
|
||
| optimize_tf_gpu(tf, K) | ||
|
|
||
|
|
||
| # convert tensor to visible image | ||
| def deprocess_image(x): | ||
| x -= x.mean() | ||
| x /= (x.std() + 1e-5) | ||
| x *= 0.1 | ||
| x += 0.5 | ||
| x = np.clip(x, 0, 1) | ||
| # x *= 255 | ||
| # x = np.clip(x, 0, 255) | ||
| # x/=255. | ||
| return x | ||
|
|
||
|
|
||
| def get_layer_type(layer): | ||
| if isinstance(layer, Conv2D): | ||
| return 'Conv2D' | ||
| elif isinstance(layer, DepthwiseConv2D): | ||
| return 'DepthwiseConv2D' | ||
| elif isinstance(layer, SeparableConv2D): | ||
| return 'SeparableConv2D' | ||
| else: | ||
| return 'Others' | ||
|
|
||
|
|
||
| # generate visualize image of conv kernel by maximize activation | ||
| # of conv kernel on random input image, with sgd on a loss function | ||
| def generate_pattern(model, layer_name, kernel_index, model_input_shape): | ||
|
|
||
| height, width = model_input_shape | ||
| # get conv kernel output of layer, and | ||
| # use mean value as loss | ||
| layer_output = model.get_layer(layer_name).output | ||
| loss = K.mean(layer_output[:, :, :, kernel_index]) | ||
| # get gradients of loss for model input | ||
| grads = K.gradients(loss, model.input)[0] | ||
| # normalize the gradients | ||
| grads /= (K.sqrt(K.mean(K.square(grads))) + 1e-5) | ||
|
|
||
| # stochastic gradient descent to maximize the loss | ||
| iterate = K.function([model.input], [loss, grads]) | ||
|
|
||
| # random input image | ||
| input_img_data = np.random.random((1, height, width, 3)) * 20 + 128. | ||
|
|
||
| # run iteration_num loops to collect pattern | ||
| step = 1. | ||
| iteration_num = 40 | ||
| for i in range(iteration_num): | ||
| loss_value, grads_value = iterate([input_img_data]) | ||
| input_img_data += grads_value * step | ||
| img = input_img_data[0] | ||
|
|
||
| return deprocess_image(img) | ||
|
|
||
|
|
||
| def convkernel_check(model_path, model_input_shape, layer_names, kernel_num, output_path): | ||
| # load model | ||
| custom_object_dict = get_custom_objects() | ||
| model = load_model(model_path, compile=False, custom_objects=custom_object_dict) | ||
| K.set_learning_phase(0) | ||
| model.summary() | ||
|
|
||
| height, width = model_input_shape | ||
|
|
||
| os.makedirs(output_path, exist_ok=True) | ||
| for layer_name in layer_names: | ||
| layer = model.get_layer(layer_name) | ||
| if get_layer_type(layer) == 'Others': | ||
| print('layer {} is not convolution type, bypass it'.format(layer_name)) | ||
| continue | ||
|
|
||
| layer_channel = layer.output.shape[-1] | ||
| if layer_channel < kernel_num: | ||
| print('layer {} does not have {} conv kernel. only pick {}'.format(layer_name, kernel_num, layer_channel)) | ||
| layer_kernel_num = layer_channel | ||
| else: | ||
| layer_kernel_num = kernel_num | ||
|
|
||
| print('Creating pattern for layer {}...'.format(layer_name)) | ||
|
|
||
| # prepare featuremap display grid array | ||
| column_num = 16 # 16 kernel pattern column per row | ||
| margin = 5 | ||
|
|
||
| row_num = layer_kernel_num // column_num | ||
| display_grid = np.zeros((height*row_num + margin*(row_num-1), width*column_num + margin*(column_num-1), 3)) | ||
|
|
||
| # fill one grid with a kernel pattern | ||
| for row in range(row_num): | ||
| for col in range(column_num): | ||
| pattern_img = generate_pattern(model, layer_name, row*column_num+col, model_input_shape) | ||
| horizontal_start = col*width + col*margin | ||
| horizontal_end = horizontal_start + width | ||
| vertical_start = row*height + row*margin | ||
| vertical_end = vertical_start + height | ||
| display_grid[vertical_start:vertical_end, horizontal_start:horizontal_end, :] = pattern_img | ||
|
|
||
| # adjust display size for grid image | ||
| row_scale = 1. / height | ||
| col_scale = 1. / width | ||
| plt.figure(figsize=(col_scale * display_grid.shape[1], | ||
| row_scale * display_grid.shape[0] + 1.8)) | ||
|
|
||
| plt.title('Conv kernel of layer '+layer_name) | ||
| plt.imshow(display_grid) | ||
|
|
||
| # save kernel pattern image & show it | ||
| pattern_file = os.path.join(output_path, layer_name+'.jpg') | ||
| plt.savefig(pattern_file, dpi=75) | ||
| #plt.show() | ||
| print('Kernel pattern of layer {} has been saved'.format(layer_name)) | ||
|
|
||
|
|
||
|
|
||
| def main(): | ||
| parser = argparse.ArgumentParser(description='check kernel pattern of specified conv layer for trained YOLO model') | ||
| parser.add_argument('--model_path', type=str, required=True, help='model file to predict') | ||
| parser.add_argument('--model_input_shape', help='model image input shape as <height>x<width>, default=%(default)s', type=str, default='416x416') | ||
| parser.add_argument('--layer_names', type=str, required=True, help='layer names to check conv kernel, separate with comma if more than one') | ||
| parser.add_argument('--kernel_num', help='conv kernel number to check, default=%(default)s', type=int, default=64) | ||
| parser.add_argument('--output_path', type=str, required=True, help='output featuremap file directory') | ||
|
|
||
| args = parser.parse_args() | ||
|
|
||
| height, width = args.model_input_shape.split('x') | ||
| model_input_shape = (int(height), int(width)) | ||
| assert (model_input_shape[0]%32 == 0 and model_input_shape[1]%32 == 0), 'model_input_shape should be multiples of 32' | ||
|
|
||
| layer_names = args.layer_names.split(',') | ||
|
|
||
| convkernel_check(args.model_path, model_input_shape, layer_names, args.kernel_num, args.output_path) | ||
|
|
||
|
|
||
| if __name__ == "__main__": | ||
| main() |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Oops, something went wrong.