dtourolle/scene-actor-extraction
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat(engine): HDF5-native galleries with embedded calibration; TensorRT backends; scene detection

Browse Source 
Gallery format switches from JSON to HDF5 exclusively (JSON read-only kept for
back-compat): save_gallery always writes HDF5, and the fitted Platt-sigmoid
calibration (a, b, valid, hash) is now embedded directly in the gallery file
instead of a sidecar .calib_cache.json — identity_matcher reads it from the
loaded gallery and writes back only when the embeddings actually changed
(hash mismatch), skipping the O(n^2) refit otherwise.

Also includes: TensorRT inference backend support (ort_backend.cpp,
trt_backend.cpp), gemm_backend improvements, TransNetV2-based scene-boundary
detection wired through frame_source/face_tracker/main, and CMake build
target updates for the new sources.

Bumps the KPN submodule to feature/persistent-pipeline-reuse (push_blocking
backpressure, node_ptr/node_stats introspection, ObjectVariantNodeWrapper for
stateful functors) — needed by the optimizer's sae_kpn Python bindings.
This commit is contained in:
dtourolle
2026-07-19 19:04:03 +02:00
parent aca6147d69
commit 41a277bc19
19 changed files with 1151 additions and 216 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/ffmpeg_decoder.hpp
+21 -3
View File
@@ -13,6 +13,7 @@ extern "C" {
#include <opencv2/core.hpp>
#include <algorithm>
#include <iostream>
#include <stdexcept>
#include <string>
@@ -35,7 +36,15 @@ extern "C" {
// Non-copyable; wrap in unique_ptr if you need to move it.
struct FFmpegDecoder {
explicit FFmpegDecoder(const std::string& path, bool use_hw = true) {
// out_scale in (0,1] downscales decoded frames (applied in the sws_scale
// colour conversion, so it's nearly free). 1.0 = native resolution.
explicit FFmpegDecoder(const std::string& path, bool use_hw = true,
float out_scale = 1.0f)
: out_scale_(out_scale > 0.f && out_scale <= 1.f ? out_scale : 1.0f)
{
// Quiet FFmpeg's own logging (e.g. the harmless "Could not dynamically
// load CUDA" emitted while probing hwaccels before VAAPI succeeds).
av_log_set_level(AV_LOG_ERROR);
if (avformat_open_input(&fmt_ctx_, path.c_str(), nullptr, nullptr) < 0)
throw std::runtime_error("[FFmpegDecoder] cannot open: " + path);
if (avformat_find_stream_info(fmt_ctx_, nullptr) < 0)
@@ -179,6 +188,7 @@ private:
AVPixelFormat hw_pix_fmt_ = AV_PIX_FMT_NONE;
int64_t last_pts_ = AV_NOPTS_VALUE;
int64_t max_forward_pts_ = AV_NOPTS_VALUE; // set after codec opens
float out_scale_ = 1.0f; // decoded-frame downscale (0,1]
int64_t to_stream_pts(double sec) const {
AVStream* s = fmt_ctx_->streams[stream_idx_];
@@ -289,13 +299,21 @@ private:
const int w = sw->width;
const int h = sw->height;
// Optional downscale, folded into the colour conversion (near-free).
// Round to even dimensions for swscale/codec friendliness.
int out_w = w, out_h = h;
if (out_scale_ < 1.0f) {
out_w = std::max(2, (static_cast<int>(w * out_scale_) / 2) * 2);
out_h = std::max(2, (static_cast<int>(h * out_scale_) / 2) * 2);
}
sws_ctx_ = sws_getCachedContext(sws_ctx_,
w, h, static_cast<AVPixelFormat>(sw->format),
w, h, AV_PIX_FMT_BGR24,
out_w, out_h, AV_PIX_FMT_BGR24,
SWS_BILINEAR, nullptr, nullptr, nullptr);
if (!sws_ctx_) return {};
cv::Mat out(h, w, CV_8UC3);
cv::Mat out(out_h, out_w, CV_8UC3);
uint8_t* dst_data[1] = { out.data };
int dst_linesize[1] = { static_cast<int>(out.step) };
sws_scale(sws_ctx_,