weight sampling example

Hi, i’ve tried to follow your weight sampling example, but i’m stuck with this error:

ValueError: Dimension 0 in both shapes must be equal, but are 3 and 324 for 'Assign_49' (op: 'Assign') with input shapes: [3,3,512,12], [324,512,3,3].

At first i thought it was because i was trying to adapt the example to SDD512, but a clean copy/paste of the original example ends up with same error. It looks like the vector is backwards, but i’m not yet fluent enough in Python/Keras/Tensorflow to actually figure where it might be happening …

I wasn’t using the original ipynb, just a local file, but should be pretty much the same as it was a plain copy/paste…

import h5py
import numpy as np
import shutil

from tensor_sampling_utils import sample_tensors

# TODO: Set the path for the source weights file you want to load.

weights_source_path = 'VGG_coco_SSD_300x300_iter_400000.h5'

# TODO: Set the path and name for the destination weights file
#       that you want to create.

# *** New class number ***
num_classes=2

weights_destination_path = 'VGG_coco_SSD_300x300_iter_400000_subsampled_'+str(num_classes)+'_classes.h5'

# Make a copy of the weights file.
shutil.copy(weights_source_path, weights_destination_path)

# Load both the source weights file and the copy we made.
# We will load the original weights file in read-only mode so that we can't mess up anything.
weights_source_file = h5py.File(weights_source_path, 'r')
weights_destination_file = h5py.File(weights_destination_path)

classifier_names = ['conv4_3_norm_mbox_conf',
                    'fc7_mbox_conf',
                    'conv6_2_mbox_conf',
                    'conv7_2_mbox_conf',
                    'conv8_2_mbox_conf',
                    'conv9_2_mbox_conf']

# *** Output the original model shape ***

conv4_3_norm_mbox_conf_kernel = weights_source_file[classifier_names[0]][classifier_names[0]]['kernel:0']
conv4_3_norm_mbox_conf_bias = weights_source_file[classifier_names[0]][classifier_names[0]]['bias:0']

print("Shape of the '{}' weights:".format(classifier_names[0]))
print()
print("kernel:\t", conv4_3_norm_mbox_conf_kernel.shape)
print("bias:\t", conv4_3_norm_mbox_conf_bias.shape)

# *** Preset paramas for the new model ***

# Number of classes that the model was trained for
n_classes_source = 81
# Number of classes we want to re-train for
# n_classes_dest = 2
# We add 1 because 'backgound' is a class too, 0 to be more precise
# classes_of_interest = list(range(n_classes_dest+1))

# The subset of classes we want to extract from our source model
# 0 SHOULD carried over to the new model.

classes_of_interest=[0,3,8]

subsampling_indices = []
for i in range(int(324/n_classes_source)):
    indices = np.array(classes_of_interest) + i * n_classes_source
    subsampling_indices.append(indices)
subsampling_indices = list(np.concatenate(subsampling_indices))

print(subsampling_indices)

for name in classifier_names:
    # Get the trained weights for this layer from the source HDF5 weights file.
    kernel = weights_source_file[name][name]['kernel:0'].value
    bias = weights_source_file[name][name]['bias:0'].value

    # Get the shape of the kernel. We're interested in sub-sampling
    # the last dimension, 'o'.
    height, width, in_channels, out_channels = kernel.shape
    
    # Compute the indices of the elements we want to sub-sample.
    # Keep in mind that each classification predictor layer predicts multiple
    # bounding boxes for every spatial location, so we want to sub-sample
    # the relevant classes for each of these boxes.
    subsampling_indices = []
    for i in range(int(out_channels/n_classes_source)):
        indices = np.array(classes_of_interest) + i * n_classes_source
        subsampling_indices.append(indices)
    subsampling_indices = list(np.concatenate(subsampling_indices))
    
    # Sub-sample the kernel and bias.
    # The `sample_tensors()` function used below provides extensive
    # documentation, so don't hesitate to read it if you want to know
    # what exactly is going on here.
    new_kernel, new_bias = sample_tensors(weights_list=[kernel, bias],
                                          sampling_instructions=[height, width, in_channels, subsampling_indices],
                                          axes=[[3]], # The one bias dimension corresponds to the last kernel dimension.
                                          init=['gaussian', 'zeros'],
                                          mean=0.0,
                                          stddev=0.005)
    
    # Delete the old weights from the destination file.
    del weights_destination_file[name][name]['kernel:0']
    del weights_destination_file[name][name]['bias:0']
    # Create new datasets for the sub-sampled weights.
    weights_destination_file[name][name].create_dataset(name='kernel:0', data=new_kernel)
    weights_destination_file[name][name].create_dataset(name='bias:0', data=new_bias)

# *** Print the new model shape ***

conv4_3_norm_mbox_conf_kernel = weights_destination_file[classifier_names[0]][classifier_names[0]]['kernel:0']
conv4_3_norm_mbox_conf_bias = weights_destination_file[classifier_names[0]][classifier_names[0]]['bias:0']

print("Shape of the '{}' weights:".format(classifier_names[0]))
print()
print("kernel:\t", conv4_3_norm_mbox_conf_kernel.shape)
print("bias:\t", conv4_3_norm_mbox_conf_bias.shape)

[pierluigiferrari]

@shivam6294 you probably forgot to set the list of classifier layers correctly for SSD512. Note that SSD512 has one more predictor layer than SSD300 (‘conv10_2_mbox_conf’). The list of classifier layers for SSD512 must be set as follows:

classifier_names = ['conv4_3_norm_mbox_conf',
                    'fc7_mbox_conf',
                    'conv6_2_mbox_conf',
                    'conv7_2_mbox_conf',
                    'conv8_2_mbox_conf',
                    'conv9_2_mbox_conf',
                    'conv10_2_mbox_conf']

In hindsight, @yf704475209 probably had the same problem.

[Alex1101a]

@shivam6294 did you solve the Problem?