marcgroef / csvm Goto Github PK

Hi,
Are there still people working on the project? Otherwise I'd like to make the repo public.

Cheers,

It would be nice (and also useful) to have to ability to export a trained configuration of the system, allowing the user to re-use, re-test, or perform further training with a system, without needing to retrain it entirely.

Hi all,

Could you copy the repo to your own account, so I can release it? :)
See your mail-box for instructions.

Cheers,

Saving seems to work right. But loading still needs to be modified to accomodate multiple codebooks

It would be nice to have a way to reflect upon the k-means progress during k means iterations

TO the extend this has/hasn't been done yet, it might be usefull to compare with the matlab code of talukdar. Perhaps checking if the input/output of feature descriptors really result in the same values.

We need a simple general train and classification functions in csvm_classifier, which also supports a validation set input. This would enable simple validation feedback while training.

Visual representations of our centroids would be nice. Preferably ordered or organised in some way that they have meaning and that preliminary conclusions or observations can be drawn from them

We need to implement the dual codebook. If possible extend it to N-codebooks. Here flexibility and cleanliness is of high importance. It will likely require a large revision of the settingsfile layout.