Permalink
Browse files

Archived TODO items from invisibleroads.com for later use

  • Loading branch information...
1 parent 292e1a6 commit b14a2d827ad986accbb7d1842b5950636a9df392 @invisibleroads committed Dec 3, 2011
Showing with 48 additions and 0 deletions.
  1. +48 −0 TODO.txt
View
@@ -0,0 +1,48 @@
+release 4.4 of remote sensing system
+ wrap eblearn c++ in swig or ctypes for python
+ make plugin for crosscompute
+
+release 5.0 of remote sensing system
+ fix core code
+ separate probability name and path from patchInformation patches
+ change terminology from probability to scan
+ integrate new EBLearn cnn code from koray
+ add time to fp go
+ make classifier scan more efficiently
+ scan entire image instead of single window at a time
+ use map-matrix to load large matrices
+
+ implement randomized connection tables
+ implement workaround for imbalanced datasets (train with half positive dataset, continue training with artificial distribution)
+ implement custom neural network architecture to combine panchromatic and multispectral matrices
+ use all training data for experiments
+ add auto calibration experiments
+ auto determine convolutional kernel sizes and subsampling ratios
+ auto determine hidden count
+ auto determine classification threshold
+
+ implement web interface
+ add queueing web service
+ implement classifiers controller for training new classifiers
+ implement simple classifier training
+ implement simple classifier delete
+
+ implement analyze image form
+ prepare some classifiers and have them listed
+ fill in processing parameters for processing results
+
+ transfer dependency from information files to model database
+ rename probability to scan
+
+release 5.1 of remote sensing system
+ write tutorials and documentation for the remote sensing system
+ write prototype that can be distributed using rabbitMQ and pyAMQP
+
+release 5.2 of remote sensing system
+ try scanning every pixel with mout
+ use part of image for training and the rest of image for testing with cross-validation
+ use the ROC curve and area under curve to evaluate performance
+ implement cluster by local binary pattern texture
+ implement captcha for houses
+ set experiments for determining optimal kernel size, subsampling ratio, hidden units
+ think of a way to more efficiently map multispectral bands with panchromatic bands

0 comments on commit b14a2d8

Please sign in to comment.