ohmgeek / faciallivenesstests Goto Github PK

Things that need doing for the solution section.

Structure:

• We need a simple structure, following all the points that we need.

Start writing the background section of the project paper, to explain the story behind the research.

Doing this now will mean less to do later on.

The image quality metric vectors produced result in fairly poor output using SVMs: Output Results:
0.4791666666666667

Therefore, it's worth trying a few different classifiers. One of these is an LDA, which is what was initially used in the paper.

Residual Network for texture classification on 2D images. This should be more effective compared to ImageNet.

I am getting exception in loading the cnn_v3.h5 file .
Exception that i am getting is below:
Unable to open file (file signature not found)

The project schedule is out of date. This should be updated to be more realistic.

Train the SVM on the data, and save the final model.
Need to do this by my next academic advisor meeting.

Model to take in a point cloud, and classify.

• Research into different existing models
• Code the one that appears to be most productive.

I tried to run live_webcam_demo.py but this error occurred:

line 223, in biqi
f = open("output_blur", "rb")
FileNotFoundError: [Errno 2] No such file or directory: 'output_blur'

Currently, only one dataset can be used at any one time.

For some models (e.g. the CNN), more data is needed, so merging multiple datasets together should help improve our model.

A combined dataset should take several instances of datasets, and the preprocess function should do the merging.

Currently, the CNN is spread out in one single file (alexnet.py).

Within the cnn namespace, we should instead create several wrapper classes/functions that allow us to build up the CNN from scratch (and to try out different modules).