Danendra Helmy Pratama, . (2026) Implementasi Mixed-Precision Quantization dengan QAT pada HDRTVNet++ untuk Aplikasi Konversi SDR ke HDR Real-Time. Skripsi thesis, Universitas Pembangunan Nasional Veteran Jakarta.
![[img]](http://repository.upnvj.ac.id/style/images/fileicons/text.png) |
Text
ABSTRAK.pdf Download (137kB) |
|
Text
AWAL.pdf Download (10MB) |
|
Text
BAB I.pdf Restricted to Repository UPNVJ Only Download (621kB) |
|
Text
BAB II.pdf Restricted to Repository UPNVJ Only Download (7MB) |
|
Text
BAB III.pdf Restricted to Repository UPNVJ Only Download (4MB) |
|
Text
BAB IV.pdf Restricted to Repository UPNVJ Only Download (32MB) |
|
Text
BAB V.pdf Download (233kB) |
|
Text
DAFTAR PUSTAKA.pdf Download (348kB) |
|
Text
RIWAYAT HIDUP.pdf Restricted to Repository staff only Download (162kB) |
|
Text
LAMPIRAN.pdf Restricted to Repository UPNVJ Only Download (14MB) |
|
Text
HASIL PLAGIARISME.pdf Restricted to Repository staff only Download (31MB) |
|
Text
ARTIKEL KI.pdf Restricted to Repository staff only Download (3MB) |
Abstract
Real-time Standard Dynamic Range (SDR) to High Dynamic Range (HDR) video conversion remains inefficient when a deep learning model is executed directly on user devices. This study aims to apply Mixed-Precision Quantization with Quantization-Aware Training (QAT) to HDRTVNet++ to improve inference speed without significantly reducing conversion quality. The method evaluates FP32, FP16, INT8 Full PTQ, INT8 Full QAT, Mixed-Precision PTQ, and Mixed-Precision QAT configurations as TensorRT engines on an NVIDIA GPU. Efficiency is evaluated through real-time playback, while accuracy is evaluated through offline benchmarking against HDR ground truth. At 1080p resolution, Mixed-Precision QAT achieves 30.89 FPS with 30.68 ms latency, while FP32 achieves 10.32 FPS with 95.49 ms latency. In terms of accuracy, Mixed-Precision QAT remains within the tolerance threshold against FP32. Therefore, Mixed-Precision QAT provides a balanced approach between inference efficiency and SDR-to-HDR real-time conversion quality.
| Item Type: | Thesis (Skripsi) |
|---|---|
| Additional Information: | [No. Panggil: 2210511039] [Pembimbing 1: Ridwan Raafi’udin] [Pembimbing 2: Muhammad Panji Muslim] [Penguji 1: Widya Cholil] [Penguji 2: I Wayan Rangga Pinastawa] |
| Uncontrolled Keywords: | HDRTVNet++, Mixed-Precision Quantization, Quantization-Aware Training, Real-Time, SDR-to-HDR Conversion. |
| Subjects: | Q Science > QA Mathematics > QA75 Electronic computers. Computer science Q Science > QA Mathematics > QA76 Computer software T Technology > T Technology (General) |
| Divisions: | Fakultas Ilmu Komputer > Program Studi Informatika (S1) |
| Depositing User: | DANENDRA HELMY PRATAMA |
| Date Deposited: | 31 Aug 2026 06:00 |
| Last Modified: | 31 Aug 2026 06:00 |
| URI: | http://repository.upnvj.ac.id/id/eprint/50954 |
Actions (login required)
 |
View Item |