curioustorvald/tsvm
1
0
Fork 0
mirror of https://github.com/curioustorvald/tsvm.git synced 2026-03-07 19:51:51 +09:00
Code Issues Packages Projects Releases Wiki Activity

TAV decoder for ffmpeg/ffplay

Browse Source
This commit is contained in:
minjaesong
2025-09-26 17:17:48 +09:00
parent efab1c3a88
commit c50d015515
4 changed files with 752 additions and 425 deletions
Download Patch File Download Diff File Expand all files Collapse all files

83
video_encoder/encoder_tav.c
View File

@@ -806,7 +806,7 @@ static void quantise_dwt_coefficients(float *coeffs, int16_t *quantised, int siz
// https://www.desmos.com/calculator/mjlpwqm8ge
// where Q=quality, x=level
static float perceptual_model3_LH(int quality, int level) {
static float perceptual_model3_LH(int quality, float level) {
float H4 = 1.2f;
float Lx = H4 - ((quality + 1.f) / 15.f) * (level - 4.f);
float Ld = (quality + 1.f) / -15.f;
@@ -824,91 +824,26 @@ static float perceptual_model3_HH(float LH, float HL) {
return (HL / LH) * 1.44f;
}
static float perceptual_model3_LL(int quality, int level) {
static float perceptual_model3_LL(int quality, float level) {
float n = perceptual_model3_LH(quality, level);
float m = perceptual_model3_LH(quality, level - 1) / n;
return n / m;
}
static float perceptual_model3_chroma_basecurve(int quality, int level) {
static float perceptual_model3_chroma_basecurve(int quality, float level) {
return 1.0f - (1.0f / (0.5f * quality * quality + 1.0f)) * (level - 4.0f); // just a line that passes (4,1)
}
// Get perceptual weight for specific subband - Data-driven model based on coefficient variance analysis
static float get_perceptual_weight_model2(int level, int subband_type, int is_chroma, int max_levels) {
// Psychovisual model based on DWT coefficient statistics and Human Visual System sensitivity
// strategy: JPEG quantisation table + real-world statistics from the encoded videos
if (!is_chroma) {
// LUMA CHANNEL: Based on statistical analysis from real video content
if (subband_type == 0) { // LL subband - contains most image energy, preserve carefully
if (level >= 6) return 0.5f; // LL6: High energy but can tolerate moderate quantisation (range up to 22K)
if (level >= 5) return 0.7f; // LL5: Good preservation
return 0.9f; // Lower LL levels: Fine preservation
} else if (subband_type == 1) { // LH subband - horizontal details (human eyes more sensitive)
if (level >= 6) return 0.8f; // LH6: Significant coefficients (max ~500), preserve well
if (level >= 5) return 1.0f; // LH5: Moderate coefficients (max ~600)
if (level >= 4) return 1.2f; // LH4: Small coefficients (max ~50)
if (level >= 3) return 1.6f; // LH3: Very small coefficients, can quantise more
if (level >= 2) return 2.0f; // LH2: Minimal impact
return 2.5f; // LH1: Least important
} else if (subband_type == 2) { // HL subband - vertical details (less sensitive due to HVS characteristics)
if (level >= 6) return 1.0f; // HL6: Can quantise more aggressively than LH6
if (level >= 5) return 1.2f; // HL5: Standard quantisation
if (level >= 4) return 1.5f; // HL4: Notable range but less critical
if (level >= 3) return 2.0f; // HL3: Can tolerate more quantisation
if (level >= 2) return 2.5f; // HL2: Less important
return 3.5f; // HL1: Most aggressive for vertical details
} else { // HH subband - diagonal details (least important for HVS)
if (level >= 6) return 1.2f; // HH6: Preserve some diagonal detail
if (level >= 5) return 1.6f; // HH5: Can quantise aggressively
if (level >= 4) return 2.0f; // HH4: Very aggressive
if (level >= 3) return 2.8f; // HH3: Minimal preservation
if (level >= 2) return 3.5f; // HH2: Maximum compression
return 5.0f; // HH1: Most aggressive quantisation
}
} else {
// CHROMA CHANNELS: Less critical for human perception, more aggressive quantisation
// strategy: mimic 4:2:2 chroma subsampling
if (subband_type == 0) { // LL chroma - still important but less than luma
return 1.0f;
if (level >= 6) return 0.8f; // Chroma LL6: Less critical than luma LL
if (level >= 5) return 0.9f;
return 1.0f;
} else if (subband_type == 1) { // LH chroma - horizontal chroma details
return 1.8f;
if (level >= 6) return 1.0f;
if (level >= 5) return 1.2f;
if (level >= 4) return 1.4f;
if (level >= 3) return 1.6f;
if (level >= 2) return 1.8f;
return 2.0f;
} else if (subband_type == 2) { // HL chroma - vertical chroma details (even less critical)
return 1.3f;
if (level >= 6) return 1.2f;
if (level >= 5) return 1.4f;
if (level >= 4) return 1.6f;
if (level >= 3) return 1.8f;
if (level >= 2) return 2.0f;
return 2.2f;
} else { // HH chroma - diagonal chroma details (most aggressive)
return 2.5f;
if (level >= 6) return 1.4f;
if (level >= 5) return 1.6f;
if (level >= 4) return 1.8f;
if (level >= 3) return 2.1f;
if (level >= 2) return 2.3f;
return 2.5f;
}
}
}
#define FOUR_PIXEL_DETAILER 0.88f
#define TWO_PIXEL_DETAILER 0.92f
// level is one-based index
static float get_perceptual_weight(tav_encoder_t *enc, int level, int subband_type, int is_chroma, int max_levels) {
static float get_perceptual_weight(tav_encoder_t *enc, int level0, int subband_type, int is_chroma, int max_levels) {
// Psychovisual model based on DWT coefficient statistics and Human Visual System sensitivity
float level = 1.0f + ((level0 - 1.0f) / (max_levels - 1.0f)) * 5.0f;
// strategy: more horizontal detail
if (!is_chroma) {
// LL subband - contains most image energy, preserve carefully
@@ -923,10 +858,10 @@ static float get_perceptual_weight(tav_encoder_t *enc, int level, int subband_ty
// HL subband - vertical details
float HL = perceptual_model3_HL(enc->quality_level, LH);
if (subband_type == 2)
return HL * (level == 2 ? TWO_PIXEL_DETAILER : level == 3 ? FOUR_PIXEL_DETAILER : 1.0f);
return HL * (2.2f >= level && level >= 1.8f ? TWO_PIXEL_DETAILER : 3.2f >= level && level >= 2.8f ? FOUR_PIXEL_DETAILER : 1.0f);
// HH subband - diagonal details
else return perceptual_model3_HH(LH, HL) * (level == 2 ? TWO_PIXEL_DETAILER : level == 3 ? FOUR_PIXEL_DETAILER : 1.0f);
else return perceptual_model3_HH(LH, HL) * (2.2f >= level && level >= 1.8f ? TWO_PIXEL_DETAILER : 3.2f >= level && level >= 2.8f ? FOUR_PIXEL_DETAILER : 1.0f);
} else {
// CHROMA CHANNELS: Less critical for human perception, more aggressive quantisation
// strategy: more horizontal detail