DoubangoTelecom compv issues

DoubangoTelecom / compv

Insanely fast Open Source Computer Vision library for ARM and x86 devices (Up to #50 times faster than OpenCV)

https://doubango.org

GNU General Public License v3.0

187 stars 39 forks source link

issues

Newest

Newest Most commented Recently updated Oldest Least commented Least recently updated

Drift in Bilinear rescale

#176 DoubangoTelecom opened 1 year ago
0
CompVMathOpSubSubMul_32f32f32f_Asm_X64_AVX requires AVX2

#175 george-hopkins closed 1 year ago
1
B not G -> change naming

#174 DoubangoTelecom closed 2 years ago
0
Change COMPV_ARM_NEON_NEQ_ZEROQ to use single instruction (instead of 2) on ARM64

#173 DoubangoTelecom closed 4 years ago
0
Add ASM code for CompVImageRemapBilinear_8u32f_Intrin

#172 DoubangoTelecom opened 5 years ago
0
Add support for POWER (ppc64el)

#171 madscientist159 opened 5 years ago
0
add support for metal (GPU, iOS, OSX)

#170 DoubangoTelecom opened 5 years ago
0
CompVMathExpExp_minpack1_64f64f_Asm_X64_AVX2 -> Intrin (VS2015) is faster than ASM

#169 DoubangoTelecom opened 6 years ago
0
Add support for LIGHTGBM

#168 DoubangoTelecom opened 6 years ago
0
Bicubic transform very slooow (OpenCV faster)

#167 DoubangoTelecom opened 6 years ago
0
Change KNN item types from "int" to "size_t"

#166 DoubangoTelecom opened 6 years ago
0
Add support for linear time morphological operator (for Toggle Mapping with large structuring element)

#165 DoubangoTelecom opened 6 years ago
0
Canny Hysteresis, use MatPtr as indexes, see UltBaseHysteresisProcessRow_8u_16mpb_Intrin_BSF_SSE2

#164 DoubangoTelecom opened 6 years ago
0
Canny, should be "colStart = (imageWidthMinus1 & -mpw))" instead of "colStart = (imageWidthMinus1 & -(mpw - 1))"

#163 DoubangoTelecom opened 6 years ago
0
ARM64 MediaPad2 -> MSER ST faster than MSER MT

#162 DoubangoTelecom opened 6 years ago
2
Use "bsf" + "blsr" to loop through a mask from _mm_movemask_epi8

#161 DoubangoTelecom opened 6 years ago
0
Call init in all newObj functions

#160 DoubangoTelecom opened 6 years ago
0
Use "CompVThreadDispatcher::dispatchDividingAcrossY" to avoid code verbosity

#159 DoubangoTelecom opened 6 years ago
0
Convolution ops could be faster

#158 DoubangoTelecom closed 6 years ago
0
Replace all "vpmovmskb + test" with "vptest"

#157 DoubangoTelecom opened 6 years ago
0
SSE41 and AVX2 no need for loading then unpack_zero for conversion -> direct memory conv exist

#156 DoubangoTelecom opened 6 years ago
0
Clean up gaussian kernel generation (see gabor for inspiration)

#155 DoubangoTelecom opened 6 years ago
0
Add support for border type in convolution

#154 DoubangoTelecom opened 6 years ago
0
CompVMat::clone doesn't work (see cloning RGB24 or YUV420P)

#153 DoubangoTelecom closed 6 years ago
1
2 remaps could be combined and done in one step

#152 DoubangoTelecom opened 6 years ago
0
Camera calibration not working with 3264x2448 images (iPhone6)

#151 DoubangoTelecom opened 6 years ago
0
Add support for "ceres-solver" a Nonlinear Least Squares Minimizer

#150 DoubangoTelecom opened 6 years ago
0
Change pattern used in camera calibration: use more connected one

#149 DoubangoTelecom opened 7 years ago
0
compv_atomic_/(inc)/(dec) -> should lead to compilation error if neither visual studio nor gcc

#148 DoubangoTelecom opened 7 years ago
0
In all ARM intrin functions, use '__compv_builtin_assume_aligned'

#147 DoubangoTelecom opened 7 years ago
0
posix_realloc_aligned -> relloc first then check if address changed

#146 DoubangoTelecom opened 7 years ago
0
Add AVX-512 implementation for histogram (gather/scatter)

#145 DoubangoTelecom opened 7 years ago
0
In arm assembler, replace "#define vecxxx vn" with "vecxxx .req vn"

#144 DoubangoTelecom opened 7 years ago
0
CompVImageScaleBilinear_Asm_XXX_YYY not faster than intrin implementations: no latency hiding technics used (rearrange instructions)

#143 DoubangoTelecom opened 7 years ago
1
yuv422family_to_y_NEON: use vld2 to unpack the luma(Y) instead of vzup

#142 DoubangoTelecom closed 7 years ago
0
Use MOVZX or MOVSX (al/ax/bx/bl/ah/bh...) to avoid "Partial register stall" issues

#141 DoubangoTelecom opened 7 years ago
1
scale bilinear, use _mm256_i32gather_epi32 to load from indices

#140 DoubangoTelecom closed 7 years ago
2
Asus PadfoneS: KHT.ST is faster than KHT.MT

#139 DoubangoTelecom closed 7 years ago
0
Add SIMD implementation for 'CompVMathEigen<double>::find2x2'

#138 DoubangoTelecom opened 7 years ago
0
Merge '__gauss_Eq15' with 'voting_Algorithm2_Gmin' and add SIMD implementation

#137 DoubangoTelecom opened 7 years ago
0
Add ASM implementation for 'CompVHoughKhtPeaks_Section3_4_VotesCount_4mpd_Intrin_NEON'

#136 DoubangoTelecom opened 7 years ago
0
Profile all AArch64 asm code with 'prfm pstl1keep' looks slower (MediaPad2)

#135 DoubangoTelecom opened 7 years ago
0
SSE and AVX supports reciprocal (_mm_rcp_ps) use such instruction for scales (1/x)

#134 DoubangoTelecom opened 7 years ago
0
Bundle (YUV -> RGB24) and (RGB24 -> HSV) and single MT instead of two MTs

#133 DoubangoTelecom opened 7 years ago
0
OpenGL rendered images looks fade

#132 DoubangoTelecom opened 7 years ago
0
Houghstd, accum increment cache-unfriendly, this is why it's slow

#131 DoubangoTelecom opened 7 years ago
0
Canny: combine with mean instead of gaussian blur to remove noise

#130 DoubangoTelecom opened 7 years ago
0
CompVCannyHysteresisRow_8mpw_Intrin_NEON is very slow

#129 DoubangoTelecom opened 7 years ago
0
No assembler functions for CompVCannyHysteresisRow_16mp_Intrin_SSE2 and CompVCannyHysteresisRow_16mp_Intrin_NEON

#128 DoubangoTelecom opened 7 years ago
0
Sobel could be done multiplication-free

#127 DoubangoTelecom opened 7 years ago
1