Refactor all keypoints tests into a single test
-
Review changes -
-
Download -
Patches
-
Plain diff
Summary
This MR refactors all keypoints tests into a single, more stable test.
Rationale
For years we endured feTvImageToFastSIFTKeyPointSetFilterSceneOutputInterestPointAscii and the like flickering our dasbhoard. They are now gone and testing strategy is stable and easier to adapt to different behavior of compilers / plateform.
Implementation Details
I removed all tests (and test code) related to Sift, SiftFast and Surf (as well as to the matching filter), and wrote a single test that generates a slightly rotated and scaled image from a reference image, and then for each aglorithm in Sift, SiftFast, and Surf do:
- Detect keypoints on both images
- Match keypoints
- Compute statistics on matching performance
- Use quality thresholds to decide if test passes or not.
Here is the ouptut of the new test:
test 559
Start 559: feTvKeyPointsAlgorithmsTest
559: Test command: /home/mp/michelj/scratch/dev/clang/otb/bin/otbDescriptorsTestDriver "otbKeyPointsAlgorithmsTest" "/work/scratch/michelj/dev/otb-data/Input/QB_TOULOUSE_RpcTag_100_100.tif"
559: Test timeout computed to be: 1500
559: 2019-01-11 16:26:48 (INFO): Default RAM limit for OTB is 128 MB
559: 2019-01-11 16:26:48 (INFO): GDAL maximum cache size is 3208 MB
559: 2019-01-11 16:26:48 (INFO): OTB will use at most 16 threads
559: 2019-01-11 16:26:49 (INFO): Loading kwl metadata from official product in file /work/scratch/michelj/dev/otb-data/Input/QB_TOULOUSE_RpcTag_100_100.tif
559: Secondary image generated by applying a rotation of 2.5 degrees and scaling of 0.989999999999999991118215802999.
559: Checking Surf implementation:
559: =============================
559: Found 405 points in reference image and 411 points in secondary image
559: Found 80 matches with 55 valid matches (tolerance of 0.5 pixels)
559: More than 15% of reference points have a match: Ok (19.753086090087890625%)
559: More than 15% of secondary points have a match: Ok (19.4647216796875%)
559: More than 65% of matches are good: Ok (68.75% at 0.5 pixel accuracy)
559: Checking Sift implementation:
559: =============================
559: Found 510 points in reference image and 486 points in secondary image
559: Found 241 matches with 217 valid matches (tolerance of 0.5 pixels)
559: More than 45% of reference points have a match: Ok (47.254901885986328125%)
559: More than 45% of secondary points have a match: Ok (49.58847808837890625%)
559: More than 85% of matches are good: Ok (90.04149627685546875% at 0.5 pixel accuracy)
559: Checking SiftFast implementation:
559: =================================
559: Found 441 points in reference image and 465 points in secondary image
559: Found 317 matches with 316 valid matches (tolerance of 0.5 pixels)
559: More than 65% of reference points have a match: Ok (71.88208770751953125%)
559: More than 65% of secondary points have a match: Ok (68.1720428466796875%)
559: More than 95% of matches are good: Ok (99.684539794921875% at 0.5 pixel accuracy)
559: -> Test EXIT SUCCESS.
559: ------------- No control baseline tests -------------
1/1 Test #559: feTvKeyPointsAlgorithmsTest ...... Passed 7.97 sec
This allowed me to remove:
- 10 test source code files
- 16 tests
- 19 baseline files from OTB-Data
Icing on the cake:
- Testing of those feature no longer relies on baselines comparisons,
- Number of test does no longer depends on OTB_USE_SIFTFAST.
Copyright
The copyright owner is CNES and has signed the ORFEO ToolBox Contributor License Agreement.
Check before merging:
- All discussions are resolved
- At least 2
- The feature branch is (reasonably) up-to-date with the base branch
- Dashboard is green
- Copyright owner has signed the ORFEO ToolBox Contributor License Agreement
Edited by Julien Michel
Merge request reports
Compare and
Compare changes
- Side-by-side
- Inline
Files
13Loading