jeremiemelo / dct_cuda Goto Github PK
View Code? Open in Web Editor NEWLicense: BSD 3-Clause "New" or "Revised" License
License: BSD 3-Clause "New" or "Revised" License
M = N = 2^x
Hi, I'm quite new to using Torch with CUDA kernels. I'm struggling to build it on Colab, do you have any advice or resources you can point me to?
Thank you so much!
Thanks for your great work!
I cloned your project ,and try to use it to do something . Then I try to generate a dct result by opencv , I used cv2.dct,then ,write the result to harddrive ad you did in script "src/dct_pytorch.py" ,but the results are quite different ,see below ,it's produced by your bin file bin/dct2d_cuda
[I] Error! Results are incorrect.
index: 0, result: 20.012664, GT: 5123.242097, scale: 0.003906
index: 1, result: -0.013377, GT: -4.842896, scale: 0.002762
index: 2, result: 0.011195, GT: 4.052967, scale: 0.002762
index: 3, result: -0.001353, GT: -0.489889, scale: 0.002762
index: 4, result: 0.001098, GT: 0.397660, scale: 0.002762
index: 5, result: 0.001302, GT: 0.471319, scale: 0.002762
index: 6, result: -0.001707, GT: -0.617927, scale: 0.002762
index: 7, result: 0.002289, GT: 0.828868, scale: 0.002762
index: 8, result: 0.009278, GT: 3.359035, scale: 0.002762
index: 9, result: -0.000801, GT: -0.289940, scale: 0.002762
index: 10, result: 0.005129, GT: 1.856900, scale: 0.002762
index: 11, result: 0.009032, GT: 3.269778, scale: 0.002762
index: 12, result: -0.000099, GT: -0.035807, scale: 0.002762
index: 13, result: 0.001421, GT: 0.514459, scale: 0.002762
index: 14, result: 0.010225, GT: 3.701966, scale: 0.002762
index: 15, result: -0.007186, GT: -2.601680, scale: 0.002762
index: 16, result: 0.000024, GT: 0.008796, scale: 0.002762
index: 17, result: -0.011907, GT: -4.310903, scale: 0.002762
index: 18, result: -0.010924, GT: -3.954782, scale: 0.002762
index: 19, result: -0.010288, GT: -3.724749, scale: 0.002762
index: 20, result: -0.005418, GT: -1.961669, scale: 0.002762
index: 21, result: 0.010174, GT: 3.683218, scale: 0.002762
index: 22, result: -0.002510, GT: -0.908892, scale: 0.002762
index: 23, result: -0.002990, GT: -1.082518, scale: 0.002762
index: 24, result: -0.010692, GT: -3.870906, scale: 0.002762
index: 25, result: -0.001337, GT: -0.483995, scale: 0.002762
index: 26, result: -0.002254, GT: -0.816180, scale: 0.002762
index: 27, result: 0.008054, GT: 2.916009, scale: 0.002762
index: 28, result: -0.003147, GT: -1.139196, scale: 0.002762
index: 29, result: 0.000077, GT: 0.027921, scale: 0.002762
index: 30, result: -0.008187, GT: -2.964190, scale: 0.002762
index: 31, result: 0.012992, GT: 4.703643, scale: 0.002762
index: 32, result: 0.006220, GT: 2.251814, scale: 0.002762
index: 33, result: 0.006298, GT: 2.280054, scale: 0.002762
index: 34, result: 0.008639, GT: 3.127580, scale: 0.002762
index: 35, result: 0.018885, GT: 6.837267, scale: 0.002762
index: 36, result: 0.001276, GT: 0.462017, scale: 0.002762
index: 37, result: -0.004175, GT: -1.511421, scale: 0.002762
index: 38, result: 0.001766, GT: 0.639508, scale: 0.002762
index: 39, result: 0.016754, GT: 6.065430, scale: 0.002762
index: 40, result: 0.008063, GT: 2.919008, scale: 0.002762
index: 41, result: -0.015651, GT: -5.666278, scale: 0.002762
index: 42, result: -0.001663, GT: -0.602239, scale: 0.002762
index: 43, result: 0.006152, GT: 2.227197, scale: 0.002762
index: 44, result: -0.001554, GT: -0.562456, scale: 0.002762
index: 45, result: 0.010431, GT: 3.776602, scale: 0.002762
index: 46, result: 0.002027, GT: 0.733966, scale: 0.002762
index: 47, result: 0.003972, GT: 1.438157, scale: 0.002762
index: 48, result: -0.007528, GT: -2.725495, scale: 0.002762
index: 49, result: -0.005630, GT: -2.038441, scale: 0.002762
index: 50, result: -0.007534, GT: -2.727680, scale: 0.002762
index: 51, result: 0.005743, GT: 2.079118, scale: 0.002762
index: 52, result: 0.004871, GT: 1.763623, scale: 0.002762
index: 53, result: -0.019159, GT: -6.936130, scale: 0.002762
index: 54, result: 0.002538, GT: 0.918893, scale: 0.002762
index: 55, result: 0.001079, GT: 0.390727, scale: 0.002762
index: 56, result: -0.004035, GT: -1.460880, scale: 0.002762
index: 57, result: -0.002547, GT: -0.922088, scale: 0.002762
index: 58, result: -0.003600, GT: -1.303184, scale: 0.002762
index: 59, result: -0.005225, GT: -1.891828, scale: 0.002762
index: 60, result: 0.001384, GT: 0.501159, scale: 0.002762
index: 61, result: 0.015481, GT: 5.604717, scale: 0.002762
index: 62, result: 0.003500, GT: 1.267265, scale: 0.002762
index: 63, result: -0.004546, GT: -1.645787, scale: 0.002762
[D] dct 2D takes 220.885 ms
[D] dct 2D (1024 * 1024) takes average 220.897 ms
I want to know the reson ,and how can I produce the same result as OpenCV do ?
Looking forward to your answers !
THX !!!!
multiplying scale in precomputeExpk
zero paddings to avoid branch divergence
in-place or out-of-place cufft, especially in idct
number of threads in idct M/2 * N/2
or M/2 * (N/2+1)
other improvements based on profiling
A declarative, efficient, and flexible JavaScript library for building user interfaces.
๐ Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. ๐๐๐
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google โค๏ธ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.