Vision Open Statistic Models

Posted on Edited on In ,

It has been quite a while that my VOSM has NOT been updated. My bad for sure. But today, I have it updated, and VOSM-0.3.5 is released. Just refer to the following 3 pages on github:

We'll still explain a bit on How to use VOSM in the following:

1. Building

1.1 Building Commands

Currently, there are 7 types of models to be built. Namely, there are 7 choices for the parameter "-t":

  • SM
  • TM
  • AM
  • IA
  • FM
  • SMLTC
  • SMNDPROFILE
1
2
3
4
5
6
7
$ testsmbuilding -o "./output" -a "./annotations/training/" -i "./images/training/" -s "../VOSM/shapeinfo/IMM/ShapeInfo.txt" -d "IMM" -c 1 -t "SM" -l 4 -p 0.95
$ testsmbuilding -o "./output" -a "./annotations/training/" -i "./images/training/" -s "../VOSM/shapeinfo/IMM/ShapeInfo.txt" -d "IMM" -c 1 -t "TM" -l 4 -p 0.95
$ testsmbuilding -o "./output" -a "./annotations/training/" -i "./images/training/" -s "../VOSM/shapeinfo/IMM/ShapeInfo.txt" -d "IMM" -c 1 -t "AM" -l 4 -p 0.95
$ testsmbuilding -o "./output" -a "./annotations/training/" -i "./images/training/" -s "../VOSM/shapeinfo/IMM/ShapeInfo.txt" -d "IMM" -c 1 -t "IA" -l 4 -p 0.95
$ testsmbuilding -o "./output" -a "./annotations/training/" -i "./images/training/" -s "../VOSM/shapeinfo/IMM/ShapeInfo.txt" -d "IMM" -c 1 -t "FM" -l 4 -p 0.95
$ testsmbuilding -o "./output" -a "./annotations/training/" -i "./images/training/" -s "../VOSM/shapeinfo/IMM/ShapeInfo.txt" -d "IMM" -c 1 -t "SMLTC" -l 4 -p 0.95
$ testsmbuilding -o "./output" -a "./annotations/training/" -i "./images/training/" -s "../VOSM/shapeinfo/IMM/ShapeInfo.txt" -d "IMM" -c 1 -t "SMNDPROFILE" -l 4 -p 0.95

1.2 Output Folders

After these 7 commands, 9 folders will be generated:

  • Point2DDistributionModel
  • ShapeModel
  • TextureModel
  • AppearanceModel
  • AAMICIA
  • AFM
  • AXM
  • ASMLTCs
  • ASMNDProfiles

1.3 Output Images

Under folder TextureModel, 3 key images are generated:

Reference.jpg edges.jpg ellipses.jpg
Reference.jpg edges.jpg ellipses

Under folder AAMICIA, another 3 key images are generated:

m_IplImageTempFace.jpg m_IplImageTempFaceX.jpg m_IplImageTempFaceY.jpg
m_IplImageTempFace.jpg - same as Reference.jpg m_IplImageTempFaceX.jpg m_IplImageTempFaceY.jpg

2. Fitting

2.1 Fitting Commands

Current VOSM supports 5 fitting methods.

  • ASM_PROFILEND
  • ASM_LTC
  • AAM_BASIC
  • AAM_CMUICIA
  • AAM_IAIA
1
2
3
4
5
$ testsmfitting -o "./output/" -t "ASM_PROFILEND" -i "./images/testing/" -a "./annotations/testing/" -d "IMM" -s true -r true
$ testsmfitting -o "./output/" -t "ASM_LTC" -i "./images/testing/" -a "./annotations/testing/" -d "IMM" -s true -r true
$ testsmfitting -o "./output/" -t "AAM_BASIC" -i "./images/testing/" -a "./annotations/testing/" -d "IMM" -s true -r true
$ testsmfitting -o "./output/" -t "AAM_CMUICIA" -i "./images/testing/" -a "./annotations/testing/" -d "IMM" -s true -r true
$ testsmfitting -o "./output/" -t "AAM_IAIA" -i "./images/testing/" -a "./annotations/testing/" -d "IMM" -s true -r true

2.2 Fitting Results

Let's just take ASM_PROFILEND as an example.

1
$ testsmfitting -o "./output/" -t "ASM_PROFILEND" -i "./images/testing/" -a "./annotations/testing/" -d "IMM" -s true -r true

All fitted images are generated under current folder, some are well fitted:

11-1m.jpg 33-4m.jpg 40-6m.jpg
11-1m.jpg 33-4m.jpg 40-6m

others are NOT well fitted:

12-3f.jpg 20-6m.jpg 23-4m.jpg
12-3f.jpg 20-6m.jpg 23-4m

2.3 Process of Fitting

The fitting process can also be recorded for each image if the parameter "-r" is enabled by -r true. Let's take a look at what's in folder 40-6m.

00.jpg 01.jpg 02.jpg
00.jpg 01.jpg 02.jpg
03.jpg 04.jpg 05.jpg
03.jpg 04.jpg 05.jpg
06.jpg 07.jpg 08.jpg
06.jpg 07.jpg 08.jpg
09.jpg 10.jpg 11.jpg
09.jpg 10.jpg 11.jpg
09.jpg 10.jpg 11.jpg
12.jpg 13.jpg 14.jpg
15.jpg 16.jpg
15.jpg 16.jpg

Clearly, the technology of pyramids is adopted during the fitting process.