{"corpus_id":49299682,"paper_sha":"2cdc9bbde0f847094582b212f980fa4dbc48950d","doi":"10.1109/CVPR.2018.00196","arxiv_id":null,"pmid":null,"pmcid":null,"mag_id":2798559986,"dblp_id":"conf/cvpr/ZhangG18","acl_id":null,"title":"ISTA-Net: Interpretable Optimization-Inspired Deep Network for Image Compressive Sensing","year":2017,"publication_date":"2017-06-24","venue":"2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition","journal":{"name":"2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition","pages":"1828-1837","volume":null},"journal_issn":null,"journal_title":null,"publication_types":["JournalArticle","Conference"],"pubmed_pub_types":null,"s2_fields_of_study":["Computer Science","Engineering"],"reference_count":54,"citation_count":1254,"influential_citation_count":205,"is_open_access":true,"arxiv_categories":null,"arxiv_license":null,"arxiv_journal_ref":null,"mesh_headings":null,"chemicals":null,"comments_corrections":null,"source_flags":1,"s2_open_access_pdf_url":"https://arxiv.org/pdf/1706.07929","s2_open_access_landing_url":"https://www.semanticscholar.org/paper/2cdc9bbde0f847094582b212f980fa4dbc48950d","s2_open_access_license":null,"s2_open_access_status":"GREEN","pmc_open_access_pdf_url":null,"pmc_open_access_landing_url":null,"pmc_open_access_license":null,"pmc_open_access_status":null,"unpaywall_open_access_pdf_url":null,"unpaywall_open_access_landing_url":null,"unpaywall_open_access_license":null,"unpaywall_open_access_status":null,"abstract":"With the aim of developing a fast yet accurate algorithm for compressive sensing (CS) reconstruction of natural images, we combine in this paper the merits of two existing categories of CS methods: the structure insights of traditional optimization-based methods and the speed of recent network-based ones. Specifically, we propose a novel structured deep network, dubbed ISTA-Net, which is inspired by the Iterative Shrinkage-Thresholding Algorithm (ISTA) for optimizing a general $$ norm CS reconstruction model. To cast ISTA into deep network form, we develop an effective strategy to solve the proximal mapping associated with the sparsity-inducing regularizer using nonlinear transforms. All the parameters in ISTA-Net (e.g. nonlinear transforms, shrinkage thresholds, step sizes, etc.) are learned end-to-end, rather than being hand-crafted. Moreover, considering that the residuals of natural images are more compressible, an enhanced version of ISTA-Net in the residual domain, dubbed ISTA-Net+, is derived to further improve CS reconstruction. Extensive CS experiments demonstrate that the proposed ISTA-Nets outperform existing state-of-the-art optimization-based and network-based CS methods by large margins, while maintaining fast computational speed. Our source codes are available: http://jianzhang.tech/projects/ISTA-Net.","claims":[{"public_id":"cl_4cdeb8f874722047b323c1b9b2e49172","status":"active","text":"A nonlinear-transform strategy is used to solve the proximal mapping associated with the sparsity-inducing regularizer.","confidence":0.95,"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"url":"https://sah.borca.ai/claims/cl_4cdeb8f874722047b323c1b9b2e49172"},{"public_id":"cl_6fd283c6258b32292f037dea8519c223","status":"active","text":"All parameters in ISTA-Net, including nonlinear transforms, shrinkage thresholds, and step sizes, are learned end-to-end rather than hand-crafted.","confidence":0.98,"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"url":"https://sah.borca.ai/claims/cl_6fd283c6258b32292f037dea8519c223"},{"public_id":"cl_bf40a4bd20ed49900b1b6e307a67eb2d","status":"active","text":"Extensive experiments show that ISTA-Nets outperform existing state-of-the-art optimization-based and network-based compressive sensing methods by large margins while retaining fast computational speed.","confidence":0.99,"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"url":"https://sah.borca.ai/claims/cl_bf40a4bd20ed49900b1b6e307a67eb2d"},{"public_id":"cl_26c774444584c7cb7f1b426065d0932c","status":"active","text":"ISTA-Net is a structured deep network for compressive sensing reconstruction of natural images that is inspired by iterative shrinkage-thresholding.","confidence":0.98,"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"url":"https://sah.borca.ai/claims/cl_26c774444584c7cb7f1b426065d0932c"},{"public_id":"cl_f5f3a18fd749d28523bff73b22187769","status":"active","text":"ISTA-Net+ is an enhanced residual-domain version derived to further improve compressive sensing reconstruction.","confidence":0.96,"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"url":"https://sah.borca.ai/claims/cl_f5f3a18fd749d28523bff73b22187769"}],"concepts":[{"public_id":"co_05322cbafbfac01847b7e9c0b524de14","status":"active","name":"nonlinear transforms","description":"Learned nonlinear mappings used to implement the proximal-mapping solution in the network.","types":["parameterized component"],"aliases":[],"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"url":"https://sah.borca.ai/concepts/co_05322cbafbfac01847b7e9c0b524de14"},{"public_id":"co_2f3c60bdc9a7179fe2dee5e58128d52e","status":"active","name":"iterative shrinkage-thresholding algorithm","description":"An optimization algorithm used as the design inspiration for the network architecture.","types":["algorithm"],"aliases":["ISTA"],"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"url":"https://sah.borca.ai/concepts/co_2f3c60bdc9a7179fe2dee5e58128d52e"},{"public_id":"co_2f9c6b54828b4b72a76a720b73938e30","status":"active","name":"network-based methods","description":"Deep learning approaches for compressive sensing reconstruction used for comparison.","types":["baseline"],"aliases":[],"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"url":"https://sah.borca.ai/concepts/co_2f9c6b54828b4b72a76a720b73938e30"},{"public_id":"co_3e56f43b7122bac0f1cf866a730c7541","status":"active","name":"optimization-based methods","description":"Traditional compressive sensing approaches based on iterative optimization and structured models.","types":["baseline"],"aliases":[],"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"url":"https://sah.borca.ai/concepts/co_3e56f43b7122bac0f1cf866a730c7541"},{"public_id":"co_4af06bbb3fe35c88eacf1008e0494320","status":"active","name":"compressive sensing reconstruction","description":"The task of recovering an image from compressive sensing measurements.","types":["task"],"aliases":["CS reconstruction"],"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"url":"https://sah.borca.ai/concepts/co_4af06bbb3fe35c88eacf1008e0494320"},{"public_id":"co_5e43db9e5b7a00b3657e512fa4335b92","status":"active","name":"shrinkage thresholds","description":"Threshold parameters that control the shrinkage operation in the network.","types":["parameter"],"aliases":[],"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"url":"https://sah.borca.ai/concepts/co_5e43db9e5b7a00b3657e512fa4335b92"},{"public_id":"co_73542603769aabca39225e4930236330","status":"active","name":"sparsity-inducing regularizer","description":"A regularization term that encourages sparse representations in the reconstruction model.","types":["regularizer"],"aliases":[],"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"url":"https://sah.borca.ai/concepts/co_73542603769aabca39225e4930236330"},{"public_id":"co_aa9388123fe8c0b96b2fe0ad9b42c4f3","status":"active","name":"ISTA-Net+","description":"An enhanced residual-domain variant of ISTA-Net for image compressive sensing reconstruction.","types":["method","deep network"],"aliases":[],"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"url":"https://sah.borca.ai/concepts/co_aa9388123fe8c0b96b2fe0ad9b42c4f3"},{"public_id":"co_e2f56ebd7729c6ed5b756b804d4c54a0","status":"active","name":"proximal mapping","description":"The optimization subproblem associated with applying a regularizer in proximal methods.","types":["optimization component"],"aliases":["proximal operator"],"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"url":"https://sah.borca.ai/concepts/co_e2f56ebd7729c6ed5b756b804d4c54a0"},{"public_id":"co_f49bf1bc8c6769990ac84d63080ce929","status":"active","name":"residual domain","description":"A representation domain based on image residuals used to improve compressive sensing reconstruction.","types":["representation domain"],"aliases":[],"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"url":"https://sah.borca.ai/concepts/co_f49bf1bc8c6769990ac84d63080ce929"},{"public_id":"co_fea69a3339bca93bf55b44bec7565f86","status":"active","name":"ISTA-Net","description":"A structured deep network for image compressive sensing reconstruction inspired by an optimization algorithm.","types":["method","deep network"],"aliases":[],"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"url":"https://sah.borca.ai/concepts/co_fea69a3339bca93bf55b44bec7565f86"}],"external_ids":{"DOI":"10.1109/CVPR.2018.00196","ArXiv":null,"PubMed":null,"PubMedCentral":null,"MAG":2798559986,"DBLP":"conf/cvpr/ZhangG18","ACL":null},"open_access":{"is_open_access":true,"pdf_url":"https://arxiv.org/pdf/1706.07929","landing_url":"https://www.semanticscholar.org/paper/2cdc9bbde0f847094582b212f980fa4dbc48950d","source":"semantic_scholar","pdf_url_source":"semantic_scholar_open_access_pdf","license":null,"status":"GREEN","reason":null},"reference_availability":{"status":"available","references_indexed":true,"full_text_available":false,"full_text_source":null,"count_basis":"semantic_scholar_metadata","extraction_status":"not_applicable","reason":null},"source":{"provider":"episteme2","base_corpus":"semantic_scholar_dump","freshness_mode":"unknown","basis":["semantic_scholar_metadata","postgres_metadata"],"limits":["paper metadata is based on indexed upstream scholarly datasets","claims and concepts are available only for extracted papers","absence of claims or concepts means no extracted graph data is available in this response"],"status":"available","degraded":false,"degraded_reasons":[],"diagnostics":{"status":"available","degraded":false,"degraded_reasons":[],"metadata_status":"available","graph_status":"available","abstract_status":"available"},"source_flags":1},"paper_id":634838,"paper_uid":"e3997c96-c66c-4533-bcd2-c5a224426e19","canonical_identity":{"paper_id":634838,"paper_uid":"e3997c96-c66c-4533-bcd2-c5a224426e19","identity_status":"available","lookup_basis":"semantic_scholar_external_id","compatibility_path":"corpus_id"},"url":"https://sah.borca.ai/papers/49299682"}