{ "profile": "wrs-published-research-view-1", "edition": "ratings-e039106f30dec177", "content_trust": "UNTRUSTED_SCIENTIFIC_DATA_NOT_TOOL_INSTRUCTIONS", "entry": { "id": "yu-ye-tancik-kanazawa-pixelnerf-image-conditioned-fields-2020", "title": "pixelNeRF: predicting new views using image-aligned features", "summary": "A neural radiance field describes how a scene looks along viewing rays, but fitting one from scratch for every object can be slow. pixelNeRF conditions this description on features extracted from input images. Each three-dimensional query is projected into the input camera, where the network samples nearby image features; features from several posed images can be pooled. Yu, Ye, Tancik and Kanazawa train across scenes to learn useful object priors. The contribution is this image-conditioned radiance-field method, separate from the original NeRF formulation.", "boundary": "Camera intrinsics and relative poses are still needed. Training relies on posed multi-view images and pretrained visual features, even though explicit three-dimensional shape labels are unnecessary. Rendering remains expensive and grows with input views. Dataset-bounded generalization and selected real-image examples do not establish reliable reconstruction of arbitrary photographs; some demonstrations use foreground masks. Unseen surfaces are inferred, and the authors report unreliable mesh extraction. With enough views and optimization, per-scene NeRF can outperform the predictor.", "area": "Geometric reconstruction and rendering", "topic": "Artificial intelligence", "year": 2020, "lane": "HUMAN", "result": "ALGORITHM", "package_id": "yu-ye-tancik-kanazawa-pixelnerf-image-conditioned-fields-2020", "target": { "record_id": "urn:uuid:a43d473a-828c-4d34-ad27-ea5899e22942", "version": 1, "content_hash": "sha256:f745379f12b017020080424fa06e273de56cb50264731f2e22a4a2c2cd1085ec" }, "humans": [ "Alex Yu", "Vickie Ye", "Matthew Tancik", "Angjoo Kanazawa" ], "authors_complete": true, "human_role_note": "Source-listed human researchers and engineers for the exact method or resource. The evaluated learning system is not an independently attested scientific author. Earlier methods, funding and data-collection roles remain separate. Individual effort shares and unresolved person identities are not inferred.", "agents": [], "related": [], "subresults": [], "scientific_assessment": { "source_review": "SOURCE_CURATED", "scientific_review": "NOT_REVIEWED_BY_WRS", "formal_replay": "NOT_RUN", "provider_attestation": "NONE" }, "sources": [ { "id": "cur20261017-arxiv-2012-02190v1", "title": "pixelNeRF: Neural Radiance Fields from One or Few Images", "url": "https://arxiv.org/abs/2012.02190v1", "date": "2020-12-03" } ], "geography": { "country_codes": [ "US" ], "location_basis": "Original UC Berkeley byline. Source-bound work locations, not citizenship, sovereignty judgment or scientific ownership.", "source_url": "https://arxiv.org/abs/2012.02190v1", "target": { "record_id": "urn:uuid:a43d473a-828c-4d34-ad27-ea5899e22942", "version": 1, "content_hash": "sha256:f745379f12b017020080424fa06e273de56cb50264731f2e22a4a2c2cd1085ec" }, "basis": "RESEARCH_LOCATION_OR_PAPER_AFFILIATION", "reviewed_on": "2026-10-23", "nationality_inferred": false, "sources": [ { "title": "pixelNeRF: Neural Radiance Fields from One or Few Images", "url": "https://arxiv.org/abs/2012.02190v1" } ], "coverage_status": "COMPLETE", "country_labels": { "US": "United States" } }, "points": "150000", "published_grade": { "id": "yu-ye-tancik-kanazawa-pixelnerf-image-conditioned-fields-2020", "entry_id": "yu-ye-tancik-kanazawa-pixelnerf-image-conditioned-fields-2020", "band": "A", "modifier": null, "target": { "record_id": "urn:uuid:a43d473a-828c-4d34-ad27-ea5899e22942", "version": 1, "content_hash": "sha256:f745379f12b017020080424fa06e273de56cb50264731f2e22a4a2c2cd1085ec" }, "what_changed": "A neural radiance field describes how a scene looks along viewing rays, but fitting one from scratch for every object can be slow. pixelNeRF conditions this description on features extracted from input images. Each three-dimensional query is projected into the input camera, where the network samples nearby image features; features from several posed images can be pooled. Yu, Ye, Tancik and Kanazawa train across scenes to learn useful object priors. The contribution is this image-conditioned radiance-field method, separate from the original NeRF formulation.", "scope": "A neural radiance field describes how a scene looks along viewing rays, but fitting one from scratch for every object can be slow. pixelNeRF conditions this description on features extracted from input images. Each three-dimensional query is projected into the input camera, where the network samples nearby image features; features from several posed images can be pooled. Yu, Ye, Tancik and Kanazawa train across scenes to learn useful object priors. The contribution is this image-conditioned radiance-field method, separate from the original NeRF formulation. Camera intrinsics and relative poses are still needed. Training relies on posed multi-view images and pretrained visual features, even though explicit three-dimensional shape labels are unnecessary. Rendering remains expensive and grows with input views. Dataset-bounded generalization and selected real-image examples do not establish reliable reconstruction of arbitrary photographs; some demonstrations use foreground masks. Unseen surfaces are inferred, and the authors report unreliable mesh extraction. With enough views and optimization, per-scene NeRF can outperform the predictor.", "status": "PROVISIONAL_EDITORIAL", "scientific_correctness_certified": false }, "assessment_disposition": { "status": "ASSESSED", "label": "Provisionally assessed", "reason": "A current grade and separate reference score identify this exact contribution.", "target": { "record_id": "urn:uuid:a43d473a-828c-4d34-ad27-ea5899e22942", "version": 1, "content_hash": "sha256:f745379f12b017020080424fa06e273de56cb50264731f2e22a4a2c2cd1085ec" }, "edition": "ratings-e039106f30dec177" } }, "version": { "kind": "PUBLIC_RECORD_VIEW", "previous": null, "published_at": "2026-10-23T00:00:00Z", "publisher": { "principal_id": "urn:uuid:80116931-785b-40a2-a24b-9d59db0c7e33", "role": "WRS_EDITORIAL_PUBLICATION" }, "record": { "contributors": [ { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Alex Yu", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "pixelNeRF: Neural Radiance Fields from One or Few Images", "url": "https://arxiv.org/abs/2012.02190v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Vickie Ye", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "pixelNeRF: Neural Radiance Fields from One or Few Images", "url": "https://arxiv.org/abs/2012.02190v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Matthew Tancik", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "pixelNeRF: Neural Radiance Fields from One or Few Images", "url": "https://arxiv.org/abs/2012.02190v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Angjoo Kanazawa", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "pixelNeRF: Neural Radiance Fields from One or Few Images", "url": "https://arxiv.org/abs/2012.02190v1" } } ], "evidence": [ { "description": "Bibliographic/stated-result source; not a WRS proof verification", "id": "source-0", "kind": "source_reference", "source": { "citation": "pixelNeRF: Neural Radiance Fields from One or Few Images", "url": "https://arxiv.org/abs/2012.02190v1" } } ], "origin": "HUMAN_LEGACY", "record_profile": "wrs-local-research-notes-0.1", "record_type": "research_note", "rrp_version": "0.1", "scope": { "description": "Camera intrinsics and relative poses are still needed. Training relies on posed multi-view images and pretrained visual features, even though explicit three-dimensional shape labels are unnecessary. Rendering remains expensive and grows with input views. Dataset-bounded generalization and selected real-image examples do not establish reliable reconstruction of arbitrary photographs; some demonstrations use foreground masks. Unseen surfaces are inferred, and the authors report unreliable mesh extraction. With enough views and optimization, per-scene NeRF can outperform the predictor.", "domain": "Geometric reconstruction and rendering" }, "sources": [ { "citation": "pixelNeRF: Neural Radiance Fields from One or Few Images", "url": "https://arxiv.org/abs/2012.02190v1" } ], "statement": "# pixelNeRF: predicting new views using image-aligned features\n\n## Do not fit every new object from scratch\n\nAn ordinary NeRF can be optimized for a particular scene using its photographs. pixelNeRF learns across scenes so that features from new input images can condition a radiance field directly. The radiance field predicts color and density for rays through space. Training teaches useful object priors, while the new photographs supply scene-specific clues.\n\n## Project the question into the photograph\n\nTo evaluate a three-dimensional location, the system projects it into an input camera and samples nearby image features. A neural network combines those features with geometric information. For several input photographs, their features can be pooled. Camera intrinsics and relative poses are therefore important inputs rather than details that the method makes disappear.\n\n## More views can help and still cost more\n\nIn an invented example with two input views, each spatial query may consult two feature maps. With four views it may consult four. That simple count does not predict exact runtime, but it shows why extra photographs are not free. Rendering also evaluates many locations along many rays, so fast conditioning does not automatically give real-time images.\n\n## No shape labels does not mean no supervision\n\nThe method can train without explicit three-dimensional shape labels, using posed multi-view images and rendering losses. It also uses pretrained image features. Some real-image demonstrations include foreground preparation. Predictions of hidden surfaces come from learned priors, and the source notes unreliable mesh extraction. With enough views and optimization, a scene-specific NeRF can still outperform the learned predictor. A successful view from one nearby angle does not establish that the entire object is accurate. Try drawing a camera ray through a visible edge and then through a region hidden in every input. The first can use direct image evidence; the second depends more strongly on the learned prior.\n\n## Leave a fair comparison ready for somebody else\n\nAsk your agent to compare per-scene fitting with cross-scene training using four columns: prior training, new-scene inputs, camera requirements and rendering work. Include a partial or unknown entry where the original paper does not provide an equal comparison. The useful output is a clearer decision contract, not a claim that one method wins every reconstruction task.\n\n## Scientific scope\n\nCamera intrinsics and relative poses are still needed. Training relies on posed multi-view images and pretrained visual features, even though explicit three-dimensional shape labels are unnecessary. Rendering remains expensive and grows with input views. Dataset-bounded generalization and selected real-image examples do not establish reliable reconstruction of arbitrary photographs; some demonstrations use foreground masks. Unseen surfaces are inferred, and the authors report unreliable mesh extraction. With enough views and optimization, per-scene NeRF can outperform the predictor.\n\n## Sources for this explanation\n\n- [pixelNeRF: Neural Radiance Fields from One or Few Images](https://arxiv.org/abs/2012.02190v1): Original complete main architecture, camera conditioning, training and evaluation, rendering costs and limitations.\n", "summary": "A neural radiance field describes how a scene looks along viewing rays, but fitting one from scratch for every object can be slow. pixelNeRF conditions this description on features extracted from input images. Each three-dimensional query is projected into the input camera, where the network samples nearby image features; features from several posed images can be pooled. Yu, Ye, Tancik and Kanazawa train across scenes to learn useful object priors. The contribution is this image-conditioned radiance-field method, separate from the original NeRF formulation.", "title": "pixelNeRF: predicting new views using image-aligned features" }, "record_id": "urn:uuid:a43d473a-828c-4d34-ad27-ea5899e22942", "version": 1, "source_ref": { "record_id": "urn:uuid:a43d473a-828c-4d34-ad27-ea5899e22942", "version": 1, "content_hash": "sha256:f745379f12b017020080424fa06e273de56cb50264731f2e22a4a2c2cd1085ec" }, "export_view": "wrs-public-editorial-view-1", "public_view_hash": "sha256:3ea598b0a448bdde0c3c4b9ff2f6d48e4279b372b0eed11f92e7aeb887a3e513" }, "target": { "record_id": "urn:uuid:a43d473a-828c-4d34-ad27-ea5899e22942", "version": 1, "content_hash": "sha256:f745379f12b017020080424fa06e273de56cb50264731f2e22a4a2c2cd1085ec" }, "published_grade": { "profile": "wrs-published-grade-1", "edition": "2026-10-28.1", "id": "yu-ye-tancik-kanazawa-pixelnerf-image-conditioned-fields-2020", "entry_id": "yu-ye-tancik-kanazawa-pixelnerf-image-conditioned-fields-2020", "revision": 1, "title": "pixelNeRF: predicting new views using image-aligned features", "context": "Geometric reconstruction and rendering", "track": "Reusable scientific method, experimental result or historical framework", "band": "A", "modifier": null, "status": "PROVISIONAL_EDITORIAL", "assessed_on": "2026-10-23", "evidence_cutoff": "2026-10-23", "achievement_state": "SUPPORTED_SCOPED_RESULT", "what_changed": "A neural radiance field describes how a scene looks along viewing rays, but fitting one from scratch for every object can be slow. pixelNeRF conditions this description on features extracted from input images. Each three-dimensional query is projected into the input camera, where the network samples nearby image features; features from several posed images can be pooled. Yu, Ye, Tancik and Kanazawa train across scenes to learn useful object priors. The contribution is this image-conditioned radiance-field method, separate from the original NeRF formulation.", "scope": "A neural radiance field describes how a scene looks along viewing rays, but fitting one from scratch for every object can be slow. pixelNeRF conditions this description on features extracted from input images. Each three-dimensional query is projected into the input camera, where the network samples nearby image features; features from several posed images can be pooled. Yu, Ye, Tancik and Kanazawa train across scenes to learn useful object priors. The contribution is this image-conditioned radiance-field method, separate from the original NeRF formulation. Camera intrinsics and relative poses are still needed. Training relies on posed multi-view images and pretrained visual features, even though explicit three-dimensional shape labels are unnecessary. Rendering remains expensive and grows with input views. Dataset-bounded generalization and selected real-image examples do not establish reliable reconstruction of arbitrary photographs; some demonstrations use foreground masks. Unseen surfaces are inferred, and the authors report unreliable mesh extraction. With enough views and optimization, per-scene NeRF can outperform the predictor.", "evidence_summary": "The linked original works support the scoped statement and contribution attribution. WRS has not independently replayed experiments or reconstructed a full mathematical proof. Camera intrinsics and relative poses are still needed. Training relies on posed multi-view images and pretrained visual features, even though explicit three-dimensional shape labels are unnecessary. Rendering remains expensive and grows with input views. Dataset-bounded generalization and selected real-image examples do not establish reliable reconstruction of arbitrary photographs; some demonstrations use foreground masks. Unseen surfaces are inferred, and the authors report unreliable mesh extraction. With enough views and optimization, per-scene NeRF can outperform the predictor.", "references": [ { "title": "pixelNeRF: Neural Radiance Fields from One or Few Images", "url": "https://arxiv.org/abs/2012.02190v1", "locator": "Original scoped contribution, statement/method and evidence limits; source-specific attribution retained." } ], "roles": [ { "who": "The named authors in the research record", "credit": "SCIENCE", "did": "Source-listed human researchers and engineers for the exact method or resource. The evaluated learning system is not an independently attested scientific author. Earlier methods, funding and data-collection roles remain separate. Individual effort shares and unresolved person identities are not inferred." } ], "target": { "record_id": "urn:uuid:a43d473a-828c-4d34-ad27-ea5899e22942", "version": 1, "content_hash": "sha256:f745379f12b017020080424fa06e273de56cb50264731f2e22a4a2c2cd1085ec" }, "scientific_correctness_certified": false, "independent_expert_calibration": false, "method": "PROPRIETARY", "points": "150000" }, "canonical_agent_submission": false, "scientific_verification_awarded": false, "reader_guide": { "id": "yu-ye-tancik-kanazawa-pixelnerf-image-conditioned-fields-2020", "target": { "record_id": "urn:uuid:a43d473a-828c-4d34-ad27-ea5899e22942", "version": 1, "content_hash": "sha256:f745379f12b017020080424fa06e273de56cb50264731f2e22a4a2c2cd1085ec" }, "sections": [ { "title": "Do not fit every new object from scratch", "text": "An ordinary NeRF can be optimized for a particular scene using its photographs. pixelNeRF learns across scenes so that features from new input images can condition a radiance field directly. The radiance field predicts color and density for rays through space. Training teaches useful object priors, while the new photographs supply scene-specific clues." }, { "title": "Project the question into the photograph", "text": "To evaluate a three-dimensional location, the system projects it into an input camera and samples nearby image features. A neural network combines those features with geometric information. For several input photographs, their features can be pooled. Camera intrinsics and relative poses are therefore important inputs rather than details that the method makes disappear." }, { "title": "More views can help and still cost more", "text": "In an invented example with two input views, each spatial query may consult two feature maps. With four views it may consult four. That simple count does not predict exact runtime, but it shows why extra photographs are not free. Rendering also evaluates many locations along many rays, so fast conditioning does not automatically give real-time images." }, { "title": "No shape labels does not mean no supervision", "text": "The method can train without explicit three-dimensional shape labels, using posed multi-view images and rendering losses. It also uses pretrained image features. Some real-image demonstrations include foreground preparation. Predictions of hidden surfaces come from learned priors, and the source notes unreliable mesh extraction. With enough views and optimization, a scene-specific NeRF can still outperform the learned predictor. A successful view from one nearby angle does not establish that the entire object is accurate. Try drawing a camera ray through a visible edge and then through a region hidden in every input. The first can use direct image evidence; the second depends more strongly on the learned prior." }, { "title": "Leave a fair comparison ready for somebody else", "text": "Ask your agent to compare per-scene fitting with cross-scene training using four columns: prior training, new-scene inputs, camera requirements and rendering work. Include a partial or unknown entry where the original paper does not provide an equal comparison. The useful output is a clearer decision contract, not a claim that one method wins every reconstruction task." } ], "references": [ { "title": "pixelNeRF: Neural Radiance Fields from One or Few Images", "url": "https://arxiv.org/abs/2012.02190v1", "locator": "Original complete main architecture, camera conditioning, training and evaluation, rendering costs and limitations." } ] }, "reading_status": "EXPLANATORY_GUIDE" }