Closed Deepank308 closed 5 years ago
@zygmuntszpak, I was searching for the standard methods for evaluations, I found out this. My doubt is : Is it a convention to use .flo file, in a standard widely accepted format(as in here), to store the ground truth flow vector or a user may directly pass the ground truth flow vector? Because I am not able to read from these .flo file in Julia.
I also referred to OpenCV contrib repository for some idea, they also end up using the .flo file for ground truth flow in the above specified format. So, I also tried to read these .flo file in Julia but couldn't, however in C++ I could.
I am still trying to successfully read the .flo files, once it's done will add a commit for the reading function.
Merging #19 into master will decrease coverage by
3.17%
. The diff coverage is50.7%
.
@@ Coverage Diff @@
## master #19 +/- ##
==========================================
- Coverage 64.8% 61.62% -3.18%
==========================================
Files 7 7
Lines 375 443 +68
==========================================
+ Hits 243 273 +30
- Misses 132 170 +38
Impacted Files | Coverage Δ | |
---|---|---|
src/ImageTracking.jl | 100% <100%> (ø) |
:arrow_up: |
src/utility.jl | 46.57% <48.52%> (-40.93%) |
:arrow_down: |
Continue to review full report at Codecov.
Legend - Click here to learn more
Δ = absolute <relative> (impact)
,ø = not affected
,? = missing data
Powered by Codecov. Last update a83b91f...be59eb4. Read the comment docs.
Merging #19 into master will decrease coverage by
1.84%
. The diff coverage is0%
.
@@ Coverage Diff @@
## master #19 +/- ##
==========================================
- Coverage 64.8% 62.95% -1.85%
==========================================
Files 7 7
Lines 375 386 +11
==========================================
Hits 243 243
- Misses 132 143 +11
Impacted Files | Coverage Δ | |
---|---|---|
src/ImageTracking.jl | 100% <ø> (ø) |
:arrow_up: |
src/utility.jl | 36.84% <0%> (-50.66%) |
:arrow_down: |
Continue to review full report at Codecov.
Legend - Click here to learn more
Δ = absolute <relative> (impact)
,ø = not affected
,? = missing data
Powered by Codecov. Last update a83b91f...71af507. Read the comment docs.
@zygmuntszpak I have put the error evaluation functions in the same utility.jl
. Or, should I keep seperately the flow visualization function and the error evaluation functions?
@zygmuntszpak I have put the error evaluation functions in the same
utility.jl
. Or, should I keep seperately the flow visualization function and the error evaluation functions?
I reckon you can put it in utility.jl
for now.
With regard to the ground truth, as far as I can tell we should allow the user to specify the ground truth flow as Array{SVector{2, Float64}, 2}
and then work on a parser that can read the .flo
file and convert it to Array{SVector{2, Float64}, 2}
.
@zygmuntszpak I haven't added the test function yet. Once, I am able to parse the .flo files I will add the test.
The last commit adds following :
1) .flo
file reading function
2) demo for error evaluation
3) test for error evaluation
4) cleans demo_franeback.jl
- moving the error evaluation part to new demo_error_evaluation.jl
file.
I have changed the name of evaluate_error function to evaluate_flow_error because there is an identically named function under test/franeback.jl
and test/lucas_kanade.jl
which was throwing error while running the tests.
@zygmuntszpak please review and suggest changes. Thank you
I'm really sorry for being so late! I was drafting my GSoC proposal.
I wonder if we shouldn't define an AbstractCoordinateConvention
and then have RasterConvention
and CartesianConvention
as subtypes instead of using strings x_y
and row_col
. For example, we currently use strings to indicate the convention convention="row_col"
.
@zygmuntszpak done
Are the tests passing for you locally? I'm not sure why all of the builds have failed on Travis and Appveyor.
Ok, can you confirm that the tests pass on your machine? If so, then I will go ahead and merge and then work on a separate pull-request to register this package etc. I just want us to be reasonably confident that we don't commit broken code to the master branch.
Yes, the tests are passing on my local machine.
Added end point error evaluation function. Currently, the function evaluates error iff ground truth flow is also provided as an array.