Improve sparse pyrLK optical flow #5771
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| float x = prevPt.x + xBase + 0.5f; | ||
| float y = prevPt.y + yBase + 0.5f; | ||
|
|
||
| I_patch[i][j] = ToFloat<T>(I(x, y)); |
There was a problem hiding this comment.
It should be I(y, x), here and below. PtrStepSz uses (y, x) coordinates order.
|
@Jet47, since you are looking at it anyway, let me assign it to you |
|
Thanks for reviewing this! I believe I've addressed all of your comments except for conversion to [0,1] range. It's not clear if you're referring to the 3 channel non texture fetch code path or the int32_t textures. |
| texture<ushort4, cudaTextureType2D, cudaReadModeNormalizedFloat> tex_I16UC4(false, cudaFilterModeLinear, cudaAddressModeClamp); | ||
|
|
||
| texture<int, cudaTextureType2D, cudaReadModeElementType> tex_I32S(false, cudaFilterModePoint, cudaAddressModeClamp); | ||
| texture<int4, cudaTextureType2D, cudaReadModeElementType> tex_I32SC4(false, cudaFilterModePoint, cudaAddressModeClamp); |
There was a problem hiding this comment.
I think you can remove textures that are not used (ushort, int and int4).
|
I've got the following compilation errors on my machine: |
|
Could you tell me a bit more about your machine? |
|
I was able to build it on my machine (Ubuntu 12.04, GCC 4.6, CUDA 7.0) after some modifications. See But now I have sanity test failures. To reproduce them download opencv_extra repository from https://github.com/Itseez/opencv_extra and run the following command: Test log: |
|
I've integrated the patch that you created. It seems to work fine so I'll push that soon. If I understand the sanity check code correctly, it's basically doing some statistics on the tracked points as detected by optical flow. If so, is it reasonable to assume there might be a bit of a difference in values due to the different texture fetch methods? IE cudaReadModeElement vs cudaReadModeNormalizedFloat. If I change the sanity test to pass in a float image and a float BGRA image then the test passes. Would that be sufficient? How were the expected values determined in the performance test? Was it generated by the cpu or gpu optical flow code? Was it verified to be correct? I ask because one of the expected min values is -0.0052967071533203125 which is outside of the image. Should I generate new truth data and more tests for the different datatypes? |
| return ret; | ||
| } | ||
|
|
||
| template <int cn, int PATCH_X, int PATCH_Y, bool calcErr, typename T = float> |
There was a problem hiding this comment.
Please do not use default template parameters for functions (typename T = float). It is C++0x feature. Just remove = float, the code should compile without it.
Yes, your solution is OK.
The expected values were generated by the same GPU code for previous run. It is a simple regression test that compares current results with previous launches. |
|
Is there anything left that I need to do for this? |
|
👍 |
|
Thank you for your contribution! |
|
@dtmoodie Could you please squash commits into one to remove useless intermediate changes? In case of problems try to follow this opencv/opencv_contrib#290 (comment). Also feel free to ask for help. |
|
Sure thing, working on squashing now. It looks like tortoiseGit doesn't want to squash due to the way I had the repo setup. I'll have the squash finished and double checked in a bit. |
… which thus allows caching of image pyramids on successive calls. Added unsigned char support for 1, 3, 4 channel images.
|
Well done! 👍 |
|
now we only need |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Modified sparse pyrlk optical flow to allow input of an image pyramid, which thus allows caching of image pyramids on successive calls.
Added uchar support for 1 and 4 channel images.
Preliminary 3 channel support, unsigned short and int support.