| Tue, 08 Sep 2015 10:58:19 +0100 |
Chris Cannam |
Fix memory leak in EM
v1.1
|
| Mon, 07 Sep 2015 12:00:22 +0100 |
Chris Cannam |
Remove unused arg
|
| Mon, 29 Jun 2015 16:22:32 +0100 |
Chris Cannam |
Initialise pitch and shift vectors to 1.0 instead of randomly -- seems to produce at least no worse results, and has the advantage of producing deterministic output
livemode
|
| Tue, 28 Apr 2015 11:24:23 +0100 |
Chris Cannam |
Merge from default branch
livemode
|
| Tue, 28 Apr 2015 09:17:12 +0100 |
Chris Cannam |
Update subrepo code
|
| Mon, 19 Jan 2015 09:33:35 +0000 |
Chris Cannam |
Update bqvec paths
livemode
|
| Wed, 06 Aug 2014 16:44:44 +0100 |
Chris Cannam |
drand48 isn't so portable and we don't really need high-quality randomness, use rand instead
|
| Mon, 04 Aug 2014 09:22:05 +0100 |
Chris Cannam |
Remove some unused member variables
|
| Thu, 22 May 2014 12:11:33 +0100 |
Chris Cannam |
Restore (and fix) the time-frequency distribution output (filtered CQ); tidy up the bottom-octave handling in draft mode
|
| Wed, 21 May 2014 15:19:02 +0100 |
Chris Cannam |
Comment
|
| Wed, 21 May 2014 14:41:53 +0100 |
Chris Cannam |
Introduce shift sparsity
|
| Tue, 20 May 2014 16:21:33 +0100 |
Chris Cannam |
Some measurements suggest we get slightly better results with more iterations, but not drastically worse with fewer -- so go lower for draft and higher for hq mode
|
| Tue, 20 May 2014 15:31:05 +0100 |
Chris Cannam |
Introduce InstrumentPack class
|
| Wed, 14 May 2014 19:38:36 +0100 |
Chris Cannam |
double -> float in EM (to test)
|
| Wed, 14 May 2014 18:09:06 +0100 |
Chris Cannam |
Some comments on things to try next...
|
| Thu, 08 May 2014 09:47:18 +0100 |
Chris Cannam |
Include in sum only the notes actually in range for an instrument (the other templates should be all zeros anyway)
|
| Wed, 07 May 2014 10:48:58 +0100 |
Chris Cannam |
Backed out changeset dfb6e5ce3eb1 -- it was arithmetically quite wrong!
bqvec-openmp
|
| Wed, 07 May 2014 10:44:18 +0100 |
Chris Cannam |
Simplify normalisation
bqvec-openmp
|
| Wed, 07 May 2014 10:38:21 +0100 |
Chris Cannam |
Eliminate a loop
bqvec-openmp
|
| Wed, 07 May 2014 09:57:19 +0100 |
Chris Cannam |
Reorder the calculations to match the series of vector operations in the most recent bqvec code, just in case it's the order of vector calculations that is saving the time rather than the avoidance of std::vector
timing
|
| Wed, 07 May 2014 09:43:15 +0100 |
Chris Cannam |
Pull out shifted templates into separate arrays with proper alignment (trading off memory occupied against getting the alignments right)
bqvec-openmp
|
| Wed, 07 May 2014 09:08:52 +0100 |
Chris Cannam |
Simplify slightly, make HQ mode the default
bqvec-openmp
|
| Wed, 07 May 2014 09:01:32 +0100 |
Chris Cannam |
Merge from default branch
bqvec-openmp
|
| Wed, 07 May 2014 08:55:44 +0100 |
Chris Cannam |
Fixed order of with_gain function in bqvec, updating correspondingly
bqvec-openmp
|
| Tue, 06 May 2014 18:55:11 +0100 |
Chris Cannam |
Add draft/intensive mode setting (determines whether to use shifts)
|
| Tue, 06 May 2014 16:05:05 +0100 |
Chris Cannam |
Update results, tidy a bit
bqvec
|
| Tue, 06 May 2014 15:12:45 +0100 |
Chris Cannam |
Ooh, that's a trappy one
bqvec
|
| Tue, 06 May 2014 14:57:09 +0100 |
Chris Cannam |
Cut allocations with temporary storage ready to swap in
bqvec
|
| Tue, 06 May 2014 14:53:11 +0100 |
Chris Cannam |
Use a couple more bqvec calls
bqvec
|
| Tue, 06 May 2014 14:45:16 +0100 |
Chris Cannam |
Use single-precision floats throughout EM code
timing
|
| Tue, 06 May 2014 14:37:42 +0100 |
Chris Cannam |
Fix silly bug
bqvec
|
| Tue, 06 May 2014 14:29:07 +0100 |
Chris Cannam |
Vectorise some calculations
bqvec
|
| Tue, 06 May 2014 14:19:19 +0100 |
Chris Cannam |
Complete using bqvec for allocation etc, but with unchanged processing logic
bqvec
|
| Tue, 06 May 2014 13:49:52 +0100 |
Chris Cannam |
First bit of bqvec adaptation
bqvec
|
| Tue, 06 May 2014 12:56:18 +0100 |
Chris Cannam |
Initialise new vectors with epsilon instead of copying and then overwriting
timing
|
| Tue, 06 May 2014 12:49:49 +0100 |
Chris Cannam |
Store templates as doubles instead of floats
timing
|
| Tue, 06 May 2014 12:43:59 +0100 |
Chris Cannam |
Reorder internal loops
timing
|
| Tue, 06 May 2014 12:39:19 +0100 |
Chris Cannam |
Merge loops
timing
|
| Tue, 06 May 2014 11:53:02 +0100 |
Chris Cannam |
Hoist variables, add consts
timing
|
| Mon, 28 Apr 2014 13:01:30 +0100 |
Chris Cannam |
Test code
|
| Tue, 08 Apr 2014 13:30:32 +0100 |
Chris Cannam |
Support shifts as an additional dimension (as in the original model). Also return velocity as well.
preshift
|
| Mon, 07 Apr 2014 17:36:40 +0100 |
Chris Cannam |
Merge from default branch
preshift
|
| Mon, 07 Apr 2014 17:34:19 +0100 |
Chris Cannam |
Fix incorrect normalisation of source activation matrix
|
| Mon, 07 Apr 2014 16:23:16 +0100 |
Chris Cannam |
Merge from default branch
preshift
|
| Mon, 07 Apr 2014 16:21:57 +0100 |
Chris Cannam |
Use supplied pitch ranges, plus subrepo update
|
| Mon, 07 Apr 2014 13:01:08 +0100 |
Chris Cannam |
Get notes out using pre-shifted templates (not working properly)
preshift
|
| Sat, 05 Apr 2014 18:10:33 +0100 |
Chris Cannam |
Sparsity parameters
|
| Sat, 05 Apr 2014 17:57:50 +0100 |
Chris Cannam |
Return an actual note transcription
|
| Fri, 04 Apr 2014 19:05:47 +0100 |
Chris Cannam |
Return pitch activation matrix
|
| Fri, 04 Apr 2014 18:18:18 +0100 |
Chris Cannam |
Calculate column sums, etc
|
| Fri, 04 Apr 2014 17:48:06 +0100 |
Chris Cannam |
Fill out the EM logic
|
| Fri, 04 Apr 2014 14:38:40 +0100 |
Chris Cannam |
More on EM
|
| Fri, 04 Apr 2014 14:28:41 +0100 |
Chris Cannam |
Start on EM class
|
