The c3d from adamcasson

I am using another trained classifier which uses c3d features and has a theano backend. If I use theano backend with your code it computes different features in fc6 layer, which are not accurate. Is there anyway I can use your code with Theano backend or are C3D weights ported in theano? Please help me out.

Problem layers shape

Hi,

First thanks for your package.

I'm encountering an issue when trying to use the pretrained C3D model. The output shape declared don't seem to systematically match the actual shape.

For example, layer pool1 is supposed to be of shape (None, 16, 56, 56, 64) but when I run the model it generate an output of shape (1, 14, 57, 57, 64)

The mistmaches aren't systematic, but can be easily checked with the following example.

Would you have any idea on why that is?

import numpy as np
from keras.models import Model
from c3d import C3D
base_model = C3D(weights='sports1M')

# input of 16-images movie
x = np.zeros((1, 16, 112, 112, 3))

# for each layer, print expected and actual output shape
for layer in base_model.layers:
    head_model = Model(inputs=base_model.input,
                       outputs=layer.output)
    print(layer.name, 
          layer.output_shape,
          head_model.predict(x).shape)

outputs:

('conv1', (None, 16, 112, 112, 64), (1, 16, 112, 112, 64))
('pool1', (None, 16, 56, 56, 64), (1, 14, 57, 57, 64))
('conv2', (None, 16, 56, 56, 128), (1, 14, 57, 57, 128))
('pool2', (None, 8, 28, 28, 128), (1, 7, 28, 28, 128))
('conv3a', (None, 8, 28, 28, 256), (1, 7, 28, 28, 256))
('conv3b', (None, 8, 28, 28, 256), (1, 7, 28, 28, 256))
('pool3', (None, 4, 14, 14, 256), (1, 3, 14, 14, 256))
('conv4a', (None, 4, 14, 14, 512), (1, 3, 14, 14, 512))
('conv4b', (None, 4, 14, 14, 512), (1, 3, 14, 14, 512))
('pool4', (None, 2, 7, 7, 512), (1, 1, 7, 7, 512))
('conv5a', (None, 2, 7, 7, 512), (1, 1, 7, 7, 512))
('conv5b', (None, 2, 7, 7, 512), (1, 1, 7, 7, 512))
('zero_padding3d_1', (None, 2, 9, 9, 512), (1, 2, 9, 9, 512))
('pool5', (None, 1, 4, 4, 512), (1, 1, 4, 4, 512))
('flatten_1', (None, 8192), (1, 8192))
('fc6', (None, 4096), (1, 4096))
('dropout_1', (None, 4096), (1, 4096))
('fc7', (None, 4096), (1, 4096))
('dropout_2', (None, 4096), (1, 4096))
('fc8', (None, 487), (1, 487))

How to generate .txt file?

Hi! I'm new to this repo and C3D. May I ask how to generate .txt file?
Thank you so much for your help!

adamcasson / c3d Goto Github PK

c3d's People

Contributors

Stargazers

Watchers

Forkers

c3d's Issues

Train

Using theano backend

Problem layers shape

How to generate .txt file?

Recommend Projects

React

Vue.js

Typescript

TensorFlow

Django

Laravel

D3

Recommend Topics

javascript

web

server

Machine learning

Visualization

Game

Recommend Org

Facebook

Microsoft

Google

Alibaba

D3

Tencent