Conversation
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
libcamera uses its own names which do not always match pixutils' names. And trying to match based on fourccs is challenging. So let's add a new field, 'libcamera_name', so that we can explicitly map between libcamera and pixutils formats. Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
The 'packed' field was a bit too ambiguous. Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
Add a helper to get the bitspp for RAW formats. Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
Disable GC for the run, do a bit more warmup, use perf_counter. Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
Start fixing the multi-plane stride issues by fixing the public API. Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
If the caller gives an integer as the stride, and the format is a multi-plane format, use similar stride extrapolation as libcamera does. Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
Many of them were quite wrong... Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
Convert a random buffer to BGR888 with each available backend (opencv, numba, numpy) and compare the outputs pairwise. Catches bugs such as channel swaps, wrong plane offsets, or off-by-one unpacking that locked per-backend SHAs in the existing test cannot detect. Comparison uses max-per-channel mean, 99th percentile, and max of the per-pixel absolute difference. Per-channel (not aggregate) mean is the key: a channel swap leaves one channel near zero while the other two diverge by ~85 on random data, which stands out even when the aggregate mean overlaps with legitimate algorithmic differences (e.g. different demosaic methods). Tolerances are split by color class (RGB exact, YUV a few LSB, RAW loose). Test discovery probes each format at 32x32 with a shared oversized buffer and emits test methods only for formats where >=2 backends accept the conversion, keeping the reported skip count at zero.
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
Allocate the output once and do the channel swap into it directly. The alpha-drop paths use a slice-copy instead of np.delete, which is simpler and avoids the strided fancy-indexing. Results are now naturally C-contiguous, and gives a nice increase in perf. Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
Shift the intermediate 16-bit array in place and then cast, instead of allocating a shifted copy and then casting. Saves one full-size buffer allocation in the RAW conversion path.
np.clip() allocates a new array by default; passing out= makes it modify the dot-product result in place. Avoids one intermediate float-array allocation per frame.
All three planes are overwritten immediately after allocation, so np.zeros() needlessly zero-fills the buffer. np.empty() skips the fill.
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
Camshark calls internal conversion functions with an integer stride, which fails now. Add a hack to convert the stride-int to a stride-tuple. Remove later. Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
In most cases this should be no-op, but it makes sure we will handle a non-contig and shaped arrays correctly too. Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
Make numpy/numba converters return None for unsupported formats so that the top-level conv.py can manage unsupported formats in one place. Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
Catching any exception hides actual issues. Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
The earlier check did the job, but it wasn't quite correct either. Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
Collaborator
|
@tomba I skimmed through the changes and that looks all sane. I think we could take the risk of just merging it :-) |
Owner
Author
Risk on! |
stefanklug
reviewed
Apr 27, 2026
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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.
Main things: