[Work Log] Implementing Likelihood server

Spent most of the day finishing coding and compiling likelihood server.

Next:

• build config file and make first run.
  • build camera files
• write matlab TCP client mex code
• send ground truth model from matlab to server
• debug:
  • check rendering: silhouettes look okay?
  • check y-axis flipping issue
  • check likelihood vs. perturbed model vs. null model vs. overexpressive model

Issue: stem radius

Until now, our model and data have been infinitesimally thin curve (medial axis).

Need to consider how to add width. Options:

1. Train?
2. Fixed? (pass as CLI parameter to server)
3. Marginalize over?
4. Optimize?

Since we're on short time, I'm leaning toward 2 for simplicity.

Building config file

Done, but need to double-check which dataset we're using in matlab at the moment.

Issues

When running batches of several datasets, need a way to ensure matlab and likelihood server are running on the same dataset.

Posted by Kyle Simek

blog comments powered by Disqus