Repository navigation
[WIP]: Add JSON model config dispatch system and model enablement fixes - #45
Open
sandeshk-intel wants to merge 3 commits into
Open
sandeshk-intel wants to merge 3 commits into
sandeshk-intel wants to merge 3 commits into
Conversation
Squashed from internal feature/sandeshk/model_config_and_enablement_innersource: - Unified JSON config dispatch for SVP/VideoProc, EDSR, CustomVSR, TSENet, RIFE, VideoSeal - HDRTVNet-LE config fixes and documentation updates - VideoProc 10-bit output / TV-range clamping fix - use-after-free fix in ff_dnn_free_model_ivsr teardown - Legacy model_type backward compatibility restored - Dockerfile/build.sh updates for patches 0005-0007 (cherry picked from commit c725e474450c4bb19c4b130842f0d4f87aeb36b7)
sandeshk-intel
requested review from
anaghdin,
marcin-le,
soopel and
surbhimadan
as code owners
August 6, 2026 12:49
Auto-generates an ivsr_ffmpeg_plugin model_config JSON from an OpenVINO IR: detects layout, nif, channel_divisor, precision, SR scale, window_type, align, and infers normalize_input/output from baked-in normalization constants in the .bin weights. Falls back to a flagged "unconfirmed" default when no signal exists in the IR. Document usage in ivsr_ffmpeg_plugin/README.md. Signed-off-by: sandeshk <sandesh.kumar.s@intel.com>
Adds patches/0008-Fix-RGB-model-10-16-bit-normalization.patch and wires it into build.sh's patch-apply sequence, right after patch 0007. pack_input_window() and unpack_output_window() in dnn_backend_ivsr.c hardcoded uint8_t pixel access and a fixed 255.0 scale for every RGB-channel model (RIFE, SPAN, EDSR, VideoSeal, HDRTVNet-LE, TSENet's sliding path). The Y-plane (VideoProc/SVP) unpack path already handled 10/16-bit samples via AVPixFmtDescriptor bit depth; the RGB path never got the equivalent treatment, so any model fed or asked to produce rgb48le/bgr48le (10/16-bit) frames had its input misread and its output corrupted (wrong byte width, wrong scale). Fix: read/write pixel components at the frame's actual bit depth, scale normalized ([0,1]) models by (2^bits-1) instead of hardcoded 255, and canonicalize raw-passthrough models ([0,255]-trained) to/from the frame's native range instead of leaving native samples unconverted. No-op for bits == 8, so existing 8-bit behavior is unchanged. Signed-off-by: sandeshk <sandesh.kumar.s@intel.com>
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
1 participant
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.
Squashed from internal feature/sandeshk/model_config_and_enablement_innersource:
(cherry picked from commit c725e474450c4bb19c4b130842f0d4f87aeb36b7)