{"about":{"non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","site":"https://codewithpapers.app","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page","syntology":{"site":"https://syntology.ai","developers":"https://syntology.ai/developers","mcp":{"server":"https://syntology.ai/mcp","transport":"streamable-http","server_card":"https://syntology.ai/.well-known/mcp/server-card.json","auth":{"type":"trial token, no account","trial_token":"https://syntology.ai/api/oauth/trial/token","method":"POST","docs":"https://syntology.ai/developers"}},"have":"https://syntology.ai/api/graph/have?x=<method, arXiv id or title> (free, answers coverage only)","paper_base":"https://syntology.ai/paper/","atlas_base":"https://app.syntology.ai/?focus="},"machine_readable":[{"url":"https://codewithpapers.app/llms.txt","what":"the machine catalog: every machine-readable file, counted"},{"url":"https://codewithpapers.app/index/manifest.json","what":"paper-to-code index by arXiv id, with Syntology's counts"},{"url":"https://codewithpapers.app/search/manifest.json","what":"site search index (titles, authors) and its files"},{"url":"https://codewithpapers.app/download","what":"bulk files: Syntology's layer, described there"},{"url":"https://codewithpapers.app/build_manifest.json","what":"the build record: inputs, counts, exclusions, probes"}]},"url":"/paper/revisiting-global-statistics-aggregation-for","title":"Improving Image Restoration by Revisiting Global Information Aggregation","arxiv_id":"2112.04491","date":"2021-12-08","proceeding":null,"authors":["Xiaojie Chu","Liangyu Chen","Chengpeng Chen","Xin Lu"],"abstract":"Global operations, such as global average pooling, are widely used in top-performance image restorers. They aggregate global information from input features along entire spatial dimensions but behave differently during training and inference in image restoration tasks: they are based on different regions, namely the cropped patches (from images) and the full-resolution images. This paper revisits global information aggregation and finds that the image-based features during inference have a different distribution than the patch-based features during training. This train-test inconsistency negatively impacts the performance of models, which is severely overlooked by previous works. To reduce the inconsistency and improve test-time performance, we propose a simple method called Test-time Local Converter (TLC). Our TLC converts global operations to local ones only during inference so that they aggregate features within local spatial regions rather than the entire large images. The proposed method can be applied to various global modules (e.g., normalization, channel and spatial attention) with negligible costs. Without the need for any fine-tuning, TLC improves state-of-the-art results on several image restoration tasks, including single-image motion deblurring, video deblurring, defocus deblurring, and image denoising. In particular, with TLC, our Restormer-Local improves the state-of-the-art result in single image deblurring from 32.92 dB to 33.57 dB on GoPro dataset. The code is available at https://github.com/megvii-research/tlc.","url_abs":"https://arxiv.org/abs/2112.04491v4","url_pdf":"https://arxiv.org/pdf/2112.04491v4.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"revisiting-global-statistics-aggregation-for","repo_url":"https://github.com/megvii-research/TLC","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"NOASSERTION"}},{"paper_slug":"revisiting-global-statistics-aggregation-for","repo_url":"https://github.com/megvii-research/NAFNet","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null},{"paper_slug":"revisiting-global-statistics-aggregation-for","repo_url":"https://github.com/setsunil/dsdnet","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"NOASSERTION"}}],"tasks":[{"task_slug":"color-image-denoising","task_name":"Color Image Denoising"},{"task_slug":"deblurring","task_name":"Deblurring"},{"task_slug":"denoising","task_name":"Denoising"},{"task_slug":"grayscale-image-denoising","task_name":"Grayscale Image Denoising"},{"task_slug":"image-deblurring","task_name":"Image Deblurring"},{"task_slug":"image-defocus-deblurring","task_name":"Image Defocus Deblurring"},{"task_slug":"image-dehazing","task_name":"Image Dehazing"},{"task_slug":"image-denoising","task_name":"Image Denoising"},{"task_slug":"image-restoration","task_name":"Image Restoration"},{"task_slug":"semantic-segmentation","task_name":"Semantic Segmentation"},{"task_slug":"single-image-deblurring","task_name":"Single Image Deblurring"},{"task_slug":"video-deblurring","task_name":"Video Deblurring"}],"methods":[{"method_slug":"average-pooling","method_name":"Average Pooling"},{"method_slug":"global-average-pooling","method_name":"Global Average Pooling"},{"method_slug":"instance-normalization","method_name":"Instance Normalization"},{"method_slug":"tlc","method_name":"TLC"}],"datasets_introduced":[],"methods_introduced":[{"slug":"tlc","name":"TLC","full_name":"Test-time Local Converter"}],"results":[{"leaderboard":"/sota/color-image-denoising-on-urban100-sigma30","task":"Color Image Denoising","dataset":"Urban100 sigma30","model":"Restormer-Local","rank_in_archive_order":1,"of":2,"metrics":{"PSNR":"33.06"},"uses_additional_data":false},{"leaderboard":"/sota/color-image-denoising-on-urban100-sigma50","task":"Color Image Denoising","dataset":"Urban100 sigma50","model":"Restormer-Local","rank_in_archive_order":3,"of":9,"metrics":{"PSNR":"30.17"},"uses_additional_data":false},{"leaderboard":"/sota/deblurring-on-gopro","task":"Deblurring","dataset":"GoPro","model":"RNN-MBP-Local","rank_in_archive_order":13,"of":56,"metrics":{"PSNR":"33.8","SSIM":"0.966"},"uses_additional_data":false},{"leaderboard":"/sota/deblurring-on-gopro","task":"Deblurring","dataset":"GoPro","model":"Restormer-Local","rank_in_archive_order":17,"of":56,"metrics":{"PSNR":"33.57","SSIM":"0.966"},"uses_additional_data":false},{"leaderboard":"/sota/deblurring-on-gopro","task":"Deblurring","dataset":"GoPro","model":"MPRNet-local","rank_in_archive_order":22,"of":56,"metrics":{"PSNR":"33.31","SSIM":"0.964"},"uses_additional_data":false},{"leaderboard":"/sota/deblurring-on-gopro","task":"Deblurring","dataset":"GoPro","model":"HINet-local","rank_in_archive_order":24,"of":56,"metrics":{"PSNR":"33.08","SSIM":"0.962"},"uses_additional_data":false},{"leaderboard":"/sota/deblurring-on-hide-trained-on-gopro","task":"Deblurring","dataset":"HIDE (trained on GOPRO)","model":"Restormer-TLC","rank_in_archive_order":12,"of":26,"metrics":{"PSNR (sRGB)":"31.49","Params (M)":"26.13","SSIM (sRGB)":"0.945"},"uses_additional_data":false},{"leaderboard":"/sota/deblurring-on-hide-trained-on-gopro","task":"Deblurring","dataset":"HIDE (trained on GOPRO)","model":"MPRNet-TLC","rank_in_archive_order":15,"of":26,"metrics":{"PSNR (sRGB)":"31.19","Params (M)":"20.1","SSIM (sRGB)":"0.942"},"uses_additional_data":false},{"leaderboard":"/sota/deblurring-on-based","task":"Deblurring","dataset":"MSU BASED","model":"MPR local","rank_in_archive_order":8,"of":11,"metrics":{"ERQAv2.0":"0.74521","LPIPS":"0.08323","PSNR":"31.65037","SSIM":"0.94542","Subjective":"0.4407","VMAF":"67.01788"},"uses_additional_data":false},{"leaderboard":"/sota/grayscale-image-denoising-on-urban100-sigma25","task":"Grayscale Image Denoising","dataset":"Urban100 sigma25","model":"Restormer-Local","rank_in_archive_order":2,"of":10,"metrics":{"PSNR":"31.55"},"uses_additional_data":false},{"leaderboard":"/sota/grayscale-image-denoising-on-urban100-sigma50","task":"Grayscale Image Denoising","dataset":"Urban100 sigma50","model":"Restormer-Local","rank_in_archive_order":2,"of":10,"metrics":{"PSNR":"28.41"},"uses_additional_data":false},{"leaderboard":"/sota/grayscale-image-denoising-on-urban100-sigma15-1","task":"Grayscale Image Denoising","dataset":"urban100 sigma15","model":"Restormer-Local","rank_in_archive_order":2,"of":3,"metrics":{"PSNR":"33.85"},"uses_additional_data":false},{"leaderboard":"/sota/image-deblurring-on-gopro","task":"Image Deblurring","dataset":"GoPro","model":"Restormer-TLC","rank_in_archive_order":20,"of":55,"metrics":{"PSNR":"33.57","Params (M)":"26.13","SSIM":"0.966"},"uses_additional_data":false},{"leaderboard":"/sota/image-deblurring-on-gopro","task":"Image Deblurring","dataset":"GoPro","model":"MPRNet-TLC","rank_in_archive_order":22,"of":55,"metrics":{"PSNR":"33.31","Params (M)":"20.1","SSIM":"0.964"},"uses_additional_data":false},{"leaderboard":"/sota/image-deblurring-on-gopro","task":"Image Deblurring","dataset":"GoPro","model":"HINet-TLC","rank_in_archive_order":28,"of":55,"metrics":{"PSNR":"33.08","SSIM":"0.962"},"uses_additional_data":false}],"syntology":{"syntology_url":"https://syntology.ai/paper/2112.04491","atlas_url":"https://app.syntology.ai/?focus=2112.04491","mcp":null,"developers":"https://syntology.ai/developers","held":true,"reason":"arxiv_id_in_atlas","checked_against":{"atlas_ids":549449,"atlas_ids_sha256":"afd72399dafd9906e9062927f70655952385a3d75488558e1f0484e950ab09b7","read_at":"2026-10-05T01:46:21Z"},"agent_calls":[{"tool":"get_citation_path","arguments":{"paper_1":"2112.04491"},"arguments_in":{"paper_2":"another paper's arXiv id or title"},"call":"get_citation_path(paper_1=\"2112.04491\", paper_2=\"…\")"},{"tool":"get_concepts_for_paper","arguments":{"arxiv_id":"2112.04491"},"call":"get_concepts_for_paper(arxiv_id=\"2112.04491\")"}],"by_repo":[{"repo":"megvii-research/TLC","kind":"official","samples":0,"ran":0,"instrument":0,"unverified":0},{"repo":"megvii-research/NAFNet","kind":"listed","samples":0,"ran":0,"instrument":0,"unverified":0,"harvested_for_other_papers":1,"harvested_for":["2204.04676"],"harvested_for_calls":[{"arxiv_id":"2204.04676","tool":"get_harvested_code_for_paper","arguments":{"arxiv_id":"2204.04676"},"call":"get_harvested_code_for_paper(arxiv_id=\"2204.04676\")"}]},{"repo":"setsunil/dsdnet","kind":"listed","samples":0,"ran":0,"instrument":0,"unverified":0}],"by_repo_harvested_for_is":"harvested_for: the arXiv ids (as Syntology's graph holds them) of every paper whose samples Syntology's graph links to this repository, in arXiv-id order, complete; the page names the first 3 and says how many more. A link only: it does not say the papers are the same work, that one copied the other, or that the code implements this paper's method; harvested_for_calls gives the get_harvested_code_for_paper call for each arXiv-shaped id among them","by_repo_is":"one row per repository in the archive's code links, Syntology's graph links or the samples; samples 0 means no sample is linked to this paper, a statement about Syntology's coverage, not about the repository: with harvested_for_other_papers present, Syntology harvested the repository for that many other papers, and without it, Syntology harvested nothing from it; placed_by_identical_code counts samples whose link records no repository, placed here because identical code was harvested from this repository for another paper; ran counts executions on a synthesized input and instrument counts failures of Syntology's instrument, not of the code"},"code_links_note":null,"arxiv_metadata":null,"syntology_extracted_results":null,"record_sha256":"ca297b2d36c681b3c12ed8cde1fd8fc49d0e128a68ba25027496dc1d89153931","record_changed_at":"2026-09-28","record_changed_at_basis":"first_hashed"}