Skip to content

Remove texture-based video processing in NvDecoder #6076

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 7 commits into from
Nov 5, 2025
Merged

Remove texture-based video processing in NvDecoder #6076

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
b893caf
Remove texture-based video processing in NvDecoder
JanuszL Oct 24, 2025
86fbad2
Review fix
JanuszL Oct 27, 2025
3cfe862
Fix usage of libswscale. Avoid intermediate quantization in ycbcr->rg…
mzient Oct 28, 2025
4a819b8
Lint.
mzient Oct 28, 2025
7c06230
Fix.
mzient Oct 28, 2025
5650689
Fix
JanuszL Oct 30, 2025
84f92be
Update DALI_EXTRA
JanuszL Nov 3, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion DALI_EXTRA_VERSION
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1 +1 @@
1ffbeaf1d085bb00f124038503508b3cb68e1a05
21dd6148f0cf4557531d54b810379c681c898e91
17 changes: 10 additions & 7 deletions dali/operators/video/color_space.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -51,18 +51,20 @@ __global__ static void VideoColorSpaceConversionKernel(
#pragma unroll
for (int j = 0; j < 2; j++) {
float cx = halfx + j * 0.5f + 0.25f;
u8vec3 yuv_val;
vec3 yuv_val;
yuv_val[0] = Y.at(ivec2{x + j, y + i}, 0, kernels::BorderClamp());

UV(&yuv_val[1], vec2(cx, cy), kernels::BorderClamp());

u8vec3 out_val;
yuv_val *= 1.0f / 255.0f;

vec3 out_val;
switch (conversion_type) {
case VIDEO_COLOR_SPACE_CONVERSION_TYPE_YUV_TO_RGB_FULL_RANGE:
out_val = dali::kernels::color::jpeg::ycbcr_to_rgb<uint8_t>(yuv_val);
out_val = dali::kernels::color::jpeg::ycbcr_to_rgb<float>(yuv_val);
break;
case VIDEO_COLOR_SPACE_CONVERSION_TYPE_YUV_TO_RGB:
out_val = dali::kernels::color::itu_r_bt_601::ycbcr_to_rgb<uint8_t>(yuv_val);
out_val = dali::kernels::color::itu_r_bt_601::ycbcr_to_rgb<float>(yuv_val);
break;
case VIDEO_COLOR_SPACE_CONVERSION_TYPE_YUV_UPSAMPLE:
out_val = yuv_val;
Expand All @@ -71,10 +73,11 @@ __global__ static void VideoColorSpaceConversionKernel(
assert(false);
}
if (normalized_range) {
output({x + j, y + i, 0}) = ConvertNorm<Out>(out_val.x);
output({x + j, y + i, 1}) = ConvertNorm<Out>(out_val.y);
output({x + j, y + i, 2}) = ConvertNorm<Out>(out_val.z);
output({x + j, y + i, 0}) = ConvertSatNorm<Out>(out_val.x);
output({x + j, y + i, 1}) = ConvertSatNorm<Out>(out_val.y);
output({x + j, y + i, 2}) = ConvertSatNorm<Out>(out_val.z);
} else {
out_val *= 255.0f;
output({x + j, y + i, 0}) = ConvertSat<Out>(out_val.x);
output({x + j, y + i, 1}) = ConvertSat<Out>(out_val.y);
output({x + j, y + i, 2}) = ConvertSat<Out>(out_val.z);
Expand Down
2 changes: 1 addition & 1 deletion dali/operators/video/frames_decoder_cpu.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -93,7 +93,7 @@ void FramesDecoderCpu::CopyToOutput(uint8_t *data) {
Width(),
Height(),
sws_output_format,
SWS_BILINEAR,
SWS_BILINEAR|SWS_FULL_CHR_H_INT|SWS_ACCURATE_RND,
Copy link
Contributor

@mzient mzient Oct 28, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This caused inaccurate results.

Copy link
Collaborator

@szkarpinski szkarpinski Oct 29, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Are there any performance implications of this change? How inaccurate were the results?

Copy link
Contributor Author

@JanuszL JanuszL Nov 5, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We didn't asses this as this is not a performance tradeoff, but something that brings the accuracy to the right level.

nullptr,
nullptr,
nullptr),
Expand Down
245 changes: 0 additions & 245 deletions dali/operators/video/legacy/reader/nvdecoder/imgproc.cu
View file
Open in desktop

This file was deleted.

33 changes: 0 additions & 33 deletions dali/operators/video/legacy/reader/nvdecoder/imgproc.h
View file
Open in desktop

This file was deleted.

Loading
Loading