Vzporedna kompresija tekstur v formatu BC1 z vmesnikom WebGPU

Cigler, Alen (author), Marolt, Matija (mentor), Lesar, Žiga (comentor)

Naloga obravnava paralelne metode kompresije tekstur v formatu BC1 (DXT1) za uporabo v realnem času, s ciljem učinkovitejšega shranjevanja in prikaza teksturiranih trikotniških modelov. Razviti so bili štirje pristopi, ki temeljijo na izbiri osnovnih barv v posameznem 4 × 4 bloku: pristop min-max, pristop z metodo PCA, naključni pristop (iskanje barv z naključnim vzorčenjem) ter pristop, ki temelji na gručenju barv. Za ovrednotenje kakovosti kompresije so uporabljene metrike MSE, PSNR, SSIM in CIEDE2000, ki merijo, v kolikšni meri rekonstruirana tekstura ohranja vizualne značilnosti izvirnika. Vse metode so implementirane s tehnologijo WebGPU in optimizirane za čim hitrejše izvajanje na grafičnih procesorjih, pri čemer so uporabljeni računski senčilniki. Izvedena je bila performančna analiza vseh pristopov ter primerjava z orodjema za kompresijo NVIDIA Texture Tools in AMD Compressonator.

• Language: Slovenian
• Keywords: kompresija tekstur, BC1, DXT1, PCA, MSE, PSNR, SSIM, CIEDE2000, CIELAB, gručenje
• Work type: Bachelor thesis/paper
• Typology: 2.11 - Undergraduate Thesis
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2025
• PID: 20.500.12556/RUL-169471
• COBISS.SI-ID: 239398403
• Publication date in RUL: 29.05.2025

Secondary language

• Language: English
• Title: Parallel compression of BC1 textures using WebGPU

Abstract

The thesis examines parallel texture compression methods in the BC1 (DXT1) format for real-time use, aiming for more efficient storage and display of textured triangular models. Four approaches were developed based on the selection of base colors in each 4 × 4 block: the min-max approach, the PCA approach, the random approach (searching for colors through random sampling), and the approach based on color clustering. To evaluate the quality of compression, metrics MSE, PSNR, SSIM, and CIEDE2000 were used, which measure the extent to which the reconstructed texture preserves the visual characteristics of the original. All methods are implemented using WebGPU technology and optimized for the fastest possible execution on graphics processors, utilizing compute shaders. A performance analysis of all approaches was conducted and a comparison with the compression tools NVIDIA Texture Tools and AMD Compressonator, was performed.

• Keywords: texture compression, BC1, DXT1, PCA, MSE, PSNR, SSIM, CIEDE2000, CIELAB, clustering