{"paper":{"title":"Real Image Denoising with Knowledge Distillation for High-Performance Mobile NPUs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"A 1.96M-parameter LiteDenoiseNet student model achieves 37.58 dB PSNR on full-resolution real image denoising benchmarks while running in 34-46 ms on mobile NPUs by leveraging NPU-compatible primitives and high-alpha knowledge distillation.","cross_cats":["cs.LG"],"primary_cat":"cs.CV","authors_text":"Asad Ahmed, Dmitry Ignatov, Faraz Kayani, Radu Timofte, Sarmad Kayani","submitted_at":"2026-05-05T12:19:30Z","abstract_excerpt":"While deep-learning-based image restoration has achieved unprecedented fidelity, deployment on mobile Neural Processing Units (NPUs) remains bottlenecked by operator incompatibility and memory-access overhead. We propose an NPU-aware hardware-algorithm co-design approach for real-world image denoising on mobile NPUs. Our approach employs a high-capacity teacher to supervise a lightweight student network specifically designed to leverage the tiled-memory architectures of modern mobile SoCs. By prioritizing NPU-native primitives -- standard 3x3 convolutions, ReLU activations, and nearest-neighbo"},"claims":{"count":3,"items":[{"kind":"strongest_claim","text":"The 1.96M-parameter student recovers 99.8% of the teacher's restoration quality via high-alpha knowledge distillation (alpha = 0.9), achieving a 21.2x parameter reduction while closing the PSNR gap from 1.63 dB to only 0.05 dB.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That restricting the student to NPU-native primitives (3x3 convolutions, ReLU, nearest-neighbor upsampling) combined with progressive context expansion up to 1024x1024 crops will preserve generalization on real-world noisy images without significant quality loss outside the specific Mobile AI 2026 benchmarks.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"A 1.96M-parameter LiteDenoiseNet student model achieves 37.58 dB PSNR on full-resolution real image denoising benchmarks while running in 34-46 ms on mobile NPUs by leveraging NPU-compatible primitives and high-alpha knowledge distillation.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"}],"snapshot_sha256":"f36e721f99bb769c0246475716a945e5ef32fc83a607ec832d3da8448df10d3b"},"source":{"id":"2605.03680","kind":"arxiv","version":1},"verdict":{"id":"986ad60d-41d1-4416-8a0a-62f5ebf607a9","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-07T17:55:39.329059Z","strongest_claim":"The 1.96M-parameter student recovers 99.8% of the teacher's restoration quality via high-alpha knowledge distillation (alpha = 0.9), achieving a 21.2x parameter reduction while closing the PSNR gap from 1.63 dB to only 0.05 dB.","one_line_summary":"A 1.96M-parameter LiteDenoiseNet student model achieves 37.58 dB PSNR on full-resolution real image denoising benchmarks while running in 34-46 ms on mobile NPUs by leveraging NPU-compatible primitives and high-alpha knowledge distillation.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That restricting the student to NPU-native primitives (3x3 convolutions, ReLU, nearest-neighbor upsampling) combined with progressive context expansion up to 1024x1024 crops will preserve generalization on real-world noisy images without significant quality loss outside the specific Mobile AI 2026 benchmarks.","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.03680/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"ai_meta_artifact","ran_at":"2026-05-20T13:36:25.249369Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-20T00:31:21.545513Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T15:06:14.581799Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"942c313fa52c2682407f1e9bb4b983f04ed10f0bc90f074dd604a4e60c26fbb0"},"references":{"count":30,"sample":[{"doi":"","year":2018,"title":"Abdelrahman Abdelhamed, Stephen Lin, and Michael S. Brown. A high-quality denoising dataset for smartphone cameras. InCVPR, pages 1692–1700, 2018. 2","work_id":"4b8cf503-8324-4766-98e7-9286ae010ac7","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2019,"title":"Abdelrahman Abdelhamed, Radu Timofte, and Michael S. Brown. NTIRE 2019 challenge on real image denoising: Methods and results. InCVPRW, pages 0–0, 2019. 2","work_id":"bb8e0d6a-6bcb-4011-8c92-9e88bed61f0d","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2020,"title":"Abdelrahman Abdelhamed, Radu Timofte, and Michael S. Brown. NTIRE 2020 challenge on real image denoising: Dataset, methods and results. InCVPRW, pages 496–497,","work_id":"677e60f8-8340-493b-b3ba-da8a234fff49","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2019,"title":"Real image denoising with feature attention","work_id":"a438b169-e5e8-41d8-8571-b778c71c47c7","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"Hinet: Half instance normalization network for image restoration","work_id":"b0e6b4c1-eb9b-4179-be1d-20b04f665bbc","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":30,"snapshot_sha256":"066c1d1e7c61f6c2d4706ed64f9f34412797e21a56352a39d42c53207cd078d6","internal_anchors":2},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}