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

more aggressive and asymmetrical Chroma compression

Browse Source
This commit is contained in:
minjaesong
2025-08-22 23:12:56 +09:00
parent b720e786a9
commit 3c135e48e0
3 changed files with 331 additions and 162 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
terranmon.txt
View File

@@ -688,7 +688,7 @@ DCT-based compression, motion compensation, and efficient temporal coding.
## Header (24 bytes)
uint8 Magic[8]: "\x1FTSVM TEV"
uint8 Version: 1
uint8 Version: 2
uint8 Flags: bit 0 = has audio
uint16 Width: video width in pixels
uint16 Height: video height in pixels
@@ -708,16 +708,19 @@ DCT-based compression, motion compensation, and efficient temporal coding.
uint32 Compressed Size
* Gzip-compressed Block Data
## Block Data (per 8x8/16x16 block)
## Block Data (per 16x16 block)
uint8 Mode: encoding mode
0x00 = SKIP (copy from previous frame)
0x01 = INTRA (DCT-coded, no prediction)
0x02 = INTER (DCT-coded with motion compensation)
0x02 = INTER (DCT-coded with motion compensation) -- currently unused due to bugs
0x03 = MOTION (motion vector only)
int16 Motion Vector X (1/4 pixel precision)
int16 Motion Vector Y (1/4 pixel precision)
int16 Motion Vector X ("capable of" 1/4 pixel precision, integer precision for now)
int16 Motion Vector Y ("capable of" 1/4 pixel precision, integer precision for now)
uint16 Coded Block Pattern (which 8x8 have non-zero coeffs)
int16 DCT Coefficients[3][64]: quantized R,G,B transform coefficients
int16[256] DCT Coefficients Y
int16[64] DCT Coefficients Co (subsampled by two)
int16[64] DCT Coefficients Cg (subsampled by two, aggressively quantised)
For SKIP and MOTION mode, DCT coefficients are filled with zero
## DCT Quantization
TEV uses 8 quality levels (0=lowest, 7=highest) with progressive quantization
@@ -725,9 +728,9 @@ tables optimized for perceptual quality. DC coefficients use fixed quantizer
of 8, while AC coefficients are quantized according to quality tables.
## Motion Compensation
- Search range: ±16 pixels
- Sub-pixel precision: 1/4 pixel
- Block size: 8x8 pixels
- Search range: ±8 pixels
- Sub-pixel precision: 1/4 pixel (again, integer precision for now)
- Block size: 16x16 pixels
- Uses Sum of Absolute Differences (SAD) for motion estimation
- Bilinear interpolation for sub-pixel motion vectors