-
Notifications
You must be signed in to change notification settings - Fork 0
/
TODO.txt
48 lines (41 loc) · 2.2 KB
/
TODO.txt
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
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