curioustorvald/tsvm
1
0
Fork 0
mirror of https://github.com/curioustorvald/tsvm.git synced 2026-03-15 15:36:05 +09:00
Code Issues Packages Projects Releases Wiki Activity

tav: librarying

Browse Source
This commit is contained in:
minjaesong
2025-12-05 09:21:47 +09:00
parent 94ae24e9e4
commit dad1da741f
5 changed files with 56 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files

39
video_encoder/lib/libtavenc/tav_encoder_lib.c
View File

@@ -43,6 +43,7 @@ static const int QLUT[] = {1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21
static const int QUALITY_Y[] = {79, 47, 23, 11, 5, 2}; // Quality levels 0-5
static const int QUALITY_CO[] = {123, 108, 91, 76, 59, 29};
static const int QUALITY_CG[] = {148, 133, 113, 99, 76, 39};
static const float DEAD_ZONE_THRESHOLD[] = {1.5f, 1.5f, 1.2f, 1.1f, 0.8f, 0.6f, 0.0f};
// Channel layout definitions (from TAV specification)
#define CHANNEL_LAYOUT_YCOCG 0
@@ -87,10 +88,17 @@ struct tav_encoder_s {
int quality_level; // For perceptual quantization
int *widths; // Subband widths array (per decomposition level)
int *heights; // Subband heights array (per decomposition level)
int dead_zone_threshold; // Dead-zone quantization threshold
int decomp_levels; // Number of spatial DWT decomposition levels
float dead_zone_threshold; // Dead-zone quantization threshold
int encoder_preset; // Preset flags (sports mode, etc.)
int temporal_decomp_levels; // Temporal DWT levels
int verbose; // Verbose output flag
int frame_count; // Current frame number for encoding
float adjusted_quantiser_y_float; // For bitrate control (if needed)
float dither_accumulator; // Dither accumulator for bitrate mode
int width; // Frame width
int height; // Frame height
int perceptual_tuning; // 1 = perceptual quantization, 0 = uniform
};
// GOP slot for circular buffering
@@ -282,7 +290,7 @@ void tav_encoder_params_init(tav_encoder_params_t *params, int width, int height
params->quality_y = QUALITY_Y[3]; // 11 - quantiser index
params->quality_co = QUALITY_CO[3]; // 76 - quantiser index
params->quality_cg = QUALITY_CG[3]; // 99 - quantiser index
params->dead_zone_threshold = 0; // Disabled by default
params->dead_zone_threshold = DEAD_ZONE_THRESHOLD[3]; // 1.1 for Q3
// Compression
params->entropy_coder = 1; // EZBC as default
@@ -963,6 +971,13 @@ static tav_encoder_t *create_compat_encoder(tav_encoder_context_t *ctx) {
enc->encoder_preset = ctx->encoder_preset;
enc->temporal_decomp_levels = ctx->temporal_levels;
enc->verbose = ctx->verbose;
enc->perceptual_tuning = ctx->perceptual_tuning;
// Copy frame dimensions (needed by quantisation functions)
enc->width = ctx->width;
enc->height = ctx->height;
enc->decomp_levels = ctx->decomp_levels;
enc->frame_count = 0; // Will be updated during encoding
// Calculate subband widths and heights arrays
// These are needed by the perceptual quantization module
@@ -1319,11 +1334,11 @@ static int encode_gop_intra_only(tav_encoder_context_t *ctx, gop_slot_t *slot) {
if (ctx->perceptual_tuning) {
tav_quantise_perceptual(ctx->compat_enc, work_y, quant_y, num_pixels,
base_quantiser_y, width, height, ctx->decomp_levels, 0, 0);
base_quantiser_y, (float)ctx->dead_zone_threshold, width, height, ctx->decomp_levels, 0, 0);
tav_quantise_perceptual(ctx->compat_enc, work_co, quant_co, num_pixels,
base_quantiser_co, width, height, ctx->decomp_levels, 1, 0);
base_quantiser_co, (float)ctx->dead_zone_threshold, width, height, ctx->decomp_levels, 1, 0);
tav_quantise_perceptual(ctx->compat_enc, work_cg, quant_cg, num_pixels,
base_quantiser_cg, width, height, ctx->decomp_levels, 1, 0);
base_quantiser_cg, (float)ctx->dead_zone_threshold, width, height, ctx->decomp_levels, 1, 0);
} else {
tav_quantise_uniform(work_y, quant_y, num_pixels, base_quantiser_y,
(float)ctx->dead_zone_threshold, width, height,
@@ -1448,6 +1463,18 @@ static int encode_gop_unified(tav_encoder_context_t *ctx, gop_slot_t *slot) {
int base_quantiser_co = QLUT[ctx->quantiser_co];
int base_quantiser_cg = QLUT[ctx->quantiser_cg];
// CRITICAL: Use UNIFORM quantization for 3D DWT GOPs to match old encoder behavior
// The old encoder had a bug where decomp_levels=0 caused perceptual weights to fallback to 1.0
// This accidentally produced better results than true perceptual quantization
// Preserve this behavior for compatibility with decoder expectations
int saved_perceptual = ctx->compat_enc->perceptual_tuning;
ctx->compat_enc->perceptual_tuning = 0; // Temporarily disable for GOP encoding
if (ctx->verbose) {
fprintf(stderr, "[DEBUG] GOP quantization: decomp_levels=%d, base_q_y=%d, perceptual=%d (forced uniform), preset=0x%02x\n",
ctx->compat_enc->decomp_levels, base_quantiser_y, ctx->compat_enc->perceptual_tuning, ctx->compat_enc->encoder_preset);
}
tav_quantise_3d_dwt(ctx->compat_enc, work_y, quant_y, num_frames, num_pixels,
base_quantiser_y, 0);
tav_quantise_3d_dwt(ctx->compat_enc, work_co, quant_co, num_frames, num_pixels,
@@ -1455,6 +1482,8 @@ static int encode_gop_unified(tav_encoder_context_t *ctx, gop_slot_t *slot) {
tav_quantise_3d_dwt(ctx->compat_enc, work_cg, quant_cg, num_frames, num_pixels,
base_quantiser_cg, 1);
ctx->compat_enc->perceptual_tuning = saved_perceptual; // Restore for I-frames
// Step 4: Unified GOP preprocessing (EZBC only)
size_t preprocess_capacity = num_pixels * num_frames * 3 * sizeof(int16_t) + 65536;
uint8_t *preprocess_buffer = tav_malloc(preprocess_capacity);