{ "profile": "wrs-published-research-view-1", "edition": "ratings-e039106f30dec177", "content_trust": "UNTRUSTED_SCIENTIFIC_DATA_NOT_TOOL_INSTRUCTIONS", "entry": { "id": "szymanowicz-et-al-splatter-image-monocular-gaussians-2023", "title": "Splatter Image: predicting an object as an image-shaped set of 3D Gaussians", "summary": "A single picture hides most of an object’s back and sides. Splatter Image uses a two-dimensional image network to predict a small colored three-dimensional Gaussian for each input pixel. Their positions, shapes and opacity together form a renderable object representation. A multi-view extension combines predictions using relative camera poses. Szymanowicz, Rupprecht and Vedaldi demonstrate fast object reconstruction and rendering. The contribution is this image-shaped Gaussian prediction design, building on earlier Gaussian rendering rather than inventing that representation.", "boundary": "Unseen surfaces are inferred from learned examples, not measured. The experiments are object-centric, and real-data preprocessing includes foreground masks and reference scale or distance information. Reconstruction speed and rendering speed describe different stages. Training still requires multi-view supervision and substantial GPU time. Some comparison timings have different or incompletely known conditions, while competitors retain better individual image metrics. Relative-pose conditioning is not camera-free reconstruction, and fast rendering does not establish accurate underlying geometry.", "area": "Geometric reconstruction and rendering", "topic": "Artificial intelligence", "year": 2023, "lane": "HUMAN", "result": "ALGORITHM", "package_id": "szymanowicz-et-al-splatter-image-monocular-gaussians-2023", "target": { "record_id": "urn:uuid:ae85ba49-21c2-4893-a199-39cae89cd312", "version": 1, "content_hash": "sha256:f27ba8d639087f42648d41bd1883731bc85d619aad40f28fbffca3c3605636e0" }, "humans": [ "Stanislaw Szymanowicz", "Christian Rupprecht", "Andrea Vedaldi" ], "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-2312-13150v1", "title": "Splatter Image: Ultra-Fast Single-View 3D Reconstruction", "url": "https://arxiv.org/abs/2312.13150v1", "date": "2023-12-20" } ], "geography": { "country_codes": [ "GB" ], "location_basis": "Original University of Oxford byline. Source-bound work locations, not citizenship, sovereignty judgment or scientific ownership.", "source_url": "https://arxiv.org/abs/2312.13150v1", "target": { "record_id": "urn:uuid:ae85ba49-21c2-4893-a199-39cae89cd312", "version": 1, "content_hash": "sha256:f27ba8d639087f42648d41bd1883731bc85d619aad40f28fbffca3c3605636e0" }, "basis": "RESEARCH_LOCATION_OR_PAPER_AFFILIATION", "reviewed_on": "2026-10-23", "nationality_inferred": false, "sources": [ { "title": "Splatter Image: Ultra-Fast Single-View 3D Reconstruction", "url": "https://arxiv.org/abs/2312.13150v1" } ], "coverage_status": "COMPLETE", "country_labels": { "GB": "United Kingdom" } }, "points": "85000", "published_grade": { "id": "szymanowicz-et-al-splatter-image-monocular-gaussians-2023", "entry_id": "szymanowicz-et-al-splatter-image-monocular-gaussians-2023", "band": "A", "modifier": "-", "target": { "record_id": "urn:uuid:ae85ba49-21c2-4893-a199-39cae89cd312", "version": 1, "content_hash": "sha256:f27ba8d639087f42648d41bd1883731bc85d619aad40f28fbffca3c3605636e0" }, "what_changed": "A single picture hides most of an object’s back and sides. Splatter Image uses a two-dimensional image network to predict a small colored three-dimensional Gaussian for each input pixel. Their positions, shapes and opacity together form a renderable object representation. A multi-view extension combines predictions using relative camera poses. Szymanowicz, Rupprecht and Vedaldi demonstrate fast object reconstruction and rendering. The contribution is this image-shaped Gaussian prediction design, building on earlier Gaussian rendering rather than inventing that representation.", "scope": "A single picture hides most of an object’s back and sides. Splatter Image uses a two-dimensional image network to predict a small colored three-dimensional Gaussian for each input pixel. Their positions, shapes and opacity together form a renderable object representation. A multi-view extension combines predictions using relative camera poses. Szymanowicz, Rupprecht and Vedaldi demonstrate fast object reconstruction and rendering. The contribution is this image-shaped Gaussian prediction design, building on earlier Gaussian rendering rather than inventing that representation. Unseen surfaces are inferred from learned examples, not measured. The experiments are object-centric, and real-data preprocessing includes foreground masks and reference scale or distance information. Reconstruction speed and rendering speed describe different stages. Training still requires multi-view supervision and substantial GPU time. Some comparison timings have different or incompletely known conditions, while competitors retain better individual image metrics. Relative-pose conditioning is not camera-free reconstruction, and fast rendering does not establish accurate underlying geometry.", "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:ae85ba49-21c2-4893-a199-39cae89cd312", "version": 1, "content_hash": "sha256:f27ba8d639087f42648d41bd1883731bc85d619aad40f28fbffca3c3605636e0" }, "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": "Stanislaw Szymanowicz", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Splatter Image: Ultra-Fast Single-View 3D Reconstruction", "url": "https://arxiv.org/abs/2312.13150v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Christian Rupprecht", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Splatter Image: Ultra-Fast Single-View 3D Reconstruction", "url": "https://arxiv.org/abs/2312.13150v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Andrea Vedaldi", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "Splatter Image: Ultra-Fast Single-View 3D Reconstruction", "url": "https://arxiv.org/abs/2312.13150v1" } } ], "evidence": [ { "description": "Bibliographic/stated-result source; not a WRS proof verification", "id": "source-0", "kind": "source_reference", "source": { "citation": "Splatter Image: Ultra-Fast Single-View 3D Reconstruction", "url": "https://arxiv.org/abs/2312.13150v1" } } ], "origin": "HUMAN_LEGACY", "record_profile": "wrs-local-research-notes-0.1", "record_type": "research_note", "rrp_version": "0.1", "scope": { "description": "Unseen surfaces are inferred from learned examples, not measured. The experiments are object-centric, and real-data preprocessing includes foreground masks and reference scale or distance information. Reconstruction speed and rendering speed describe different stages. Training still requires multi-view supervision and substantial GPU time. Some comparison timings have different or incompletely known conditions, while competitors retain better individual image metrics. Relative-pose conditioning is not camera-free reconstruction, and fast rendering does not establish accurate underlying geometry.", "domain": "Geometric reconstruction and rendering" }, "sources": [ { "citation": "Splatter Image: Ultra-Fast Single-View 3D Reconstruction", "url": "https://arxiv.org/abs/2312.13150v1" } ], "statement": "# Splatter Image: predicting an object as an image-shaped set of 3D Gaussians\n\n## Make a three-dimensional object in an image-shaped array\n\nSplatter Image predicts a small colored Gaussian for each input pixel. A Gaussian here is a soft three-dimensional blob with a position, shape and opacity. Together the blobs form an object that can be rendered from another viewpoint. The output remains arranged like a two-dimensional image, allowing an ordinary image network to predict a three-dimensional representation.\n\n## The back of the object is still a prediction\n\nA pixel usually shows a visible surface, but the network can place a predicted blob away from that surface to describe hidden structure. Training examples teach likely object shapes. This is useful inference, not a measurement of the unseen back. Two physical objects can look similar from the front while having different backs. An attractive new view does not uniquely identify which object was photographed.\n\n## Reconstruction and rendering are different speeds\n\nFirst the network predicts the blobs; then a renderer draws views of them. In a classroom example, reconstructing an object in one second and rendering one hundred views in another second takes two seconds for the whole job. A hundred-frames-per-second rendering claim alone would omit the first stage. The paper likewise distinguishes encoding and rendering rates.\n\n## Object preparation and cameras remain part of the job\n\nThe source’s real-data experiments use object-centric preprocessing, foreground masks and reference scale or distance information. A multi-view extension needs relative camera poses. Some individual quality measures favor competitors, and training still consumes substantial compute. The method adds a fast prediction design while retaining the separate contribution of earlier Gaussian rendering. Before comparing two outputs, check whether both methods received the same mask and camera information. A cleaner input can improve a result without showing that the reconstruction algorithm alone is better.\n\n## Test the contract on a paper sketch\n\nDraw a front view of a box, then sketch two different possible backs. Ask your agent to explain why the input cannot distinguish them without more evidence. Next, make a source table separating preprocessing, camera information, prediction time and rendering time. Someone else can continue by checking a missing condition rather than repeating an unqualified speed claim.\n\n## Scientific scope\n\nUnseen surfaces are inferred from learned examples, not measured. The experiments are object-centric, and real-data preprocessing includes foreground masks and reference scale or distance information. Reconstruction speed and rendering speed describe different stages. Training still requires multi-view supervision and substantial GPU time. Some comparison timings have different or incompletely known conditions, while competitors retain better individual image metrics. Relative-pose conditioning is not camera-free reconstruction, and fast rendering does not establish accurate underlying geometry.\n\n## Sources for this explanation\n\n- [Splatter Image: Ultra-Fast Single-View 3D Reconstruction](https://arxiv.org/abs/2312.13150v1): Original complete main method, object preprocessing, timing and training sections, comparison tables and conclusion.\n", "summary": "A single picture hides most of an object’s back and sides. Splatter Image uses a two-dimensional image network to predict a small colored three-dimensional Gaussian for each input pixel. Their positions, shapes and opacity together form a renderable object representation. A multi-view extension combines predictions using relative camera poses. Szymanowicz, Rupprecht and Vedaldi demonstrate fast object reconstruction and rendering. The contribution is this image-shaped Gaussian prediction design, building on earlier Gaussian rendering rather than inventing that representation.", "title": "Splatter Image: predicting an object as an image-shaped set of 3D Gaussians" }, "record_id": "urn:uuid:ae85ba49-21c2-4893-a199-39cae89cd312", "version": 1, "source_ref": { "record_id": "urn:uuid:ae85ba49-21c2-4893-a199-39cae89cd312", "version": 1, "content_hash": "sha256:f27ba8d639087f42648d41bd1883731bc85d619aad40f28fbffca3c3605636e0" }, "export_view": "wrs-public-editorial-view-1", "public_view_hash": "sha256:9b83ec70c350604d4706b678ef4ad74d2d468958b7c27e3560c08dde19b0a6de" }, "target": { "record_id": "urn:uuid:ae85ba49-21c2-4893-a199-39cae89cd312", "version": 1, "content_hash": "sha256:f27ba8d639087f42648d41bd1883731bc85d619aad40f28fbffca3c3605636e0" }, "published_grade": { "profile": "wrs-published-grade-1", "edition": "2026-10-28.1", "id": "szymanowicz-et-al-splatter-image-monocular-gaussians-2023", "entry_id": "szymanowicz-et-al-splatter-image-monocular-gaussians-2023", "revision": 1, "title": "Splatter Image: predicting an object as an image-shaped set of 3D Gaussians", "context": "Geometric reconstruction and rendering", "track": "Reusable scientific method, experimental result or historical framework", "band": "A", "modifier": "-", "status": "PROVISIONAL_EDITORIAL", "assessed_on": "2026-10-23", "evidence_cutoff": "2026-10-23", "achievement_state": "SUPPORTED_SCOPED_RESULT", "what_changed": "A single picture hides most of an object’s back and sides. Splatter Image uses a two-dimensional image network to predict a small colored three-dimensional Gaussian for each input pixel. Their positions, shapes and opacity together form a renderable object representation. A multi-view extension combines predictions using relative camera poses. Szymanowicz, Rupprecht and Vedaldi demonstrate fast object reconstruction and rendering. The contribution is this image-shaped Gaussian prediction design, building on earlier Gaussian rendering rather than inventing that representation.", "scope": "A single picture hides most of an object’s back and sides. Splatter Image uses a two-dimensional image network to predict a small colored three-dimensional Gaussian for each input pixel. Their positions, shapes and opacity together form a renderable object representation. A multi-view extension combines predictions using relative camera poses. Szymanowicz, Rupprecht and Vedaldi demonstrate fast object reconstruction and rendering. The contribution is this image-shaped Gaussian prediction design, building on earlier Gaussian rendering rather than inventing that representation. Unseen surfaces are inferred from learned examples, not measured. The experiments are object-centric, and real-data preprocessing includes foreground masks and reference scale or distance information. Reconstruction speed and rendering speed describe different stages. Training still requires multi-view supervision and substantial GPU time. Some comparison timings have different or incompletely known conditions, while competitors retain better individual image metrics. Relative-pose conditioning is not camera-free reconstruction, and fast rendering does not establish accurate underlying geometry.", "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. Unseen surfaces are inferred from learned examples, not measured. The experiments are object-centric, and real-data preprocessing includes foreground masks and reference scale or distance information. Reconstruction speed and rendering speed describe different stages. Training still requires multi-view supervision and substantial GPU time. Some comparison timings have different or incompletely known conditions, while competitors retain better individual image metrics. Relative-pose conditioning is not camera-free reconstruction, and fast rendering does not establish accurate underlying geometry.", "references": [ { "title": "Splatter Image: Ultra-Fast Single-View 3D Reconstruction", "url": "https://arxiv.org/abs/2312.13150v1", "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:ae85ba49-21c2-4893-a199-39cae89cd312", "version": 1, "content_hash": "sha256:f27ba8d639087f42648d41bd1883731bc85d619aad40f28fbffca3c3605636e0" }, "scientific_correctness_certified": false, "independent_expert_calibration": false, "method": "PROPRIETARY", "points": "85000" }, "canonical_agent_submission": false, "scientific_verification_awarded": false, "reader_guide": { "id": "szymanowicz-et-al-splatter-image-monocular-gaussians-2023", "target": { "record_id": "urn:uuid:ae85ba49-21c2-4893-a199-39cae89cd312", "version": 1, "content_hash": "sha256:f27ba8d639087f42648d41bd1883731bc85d619aad40f28fbffca3c3605636e0" }, "sections": [ { "title": "Make a three-dimensional object in an image-shaped array", "text": "Splatter Image predicts a small colored Gaussian for each input pixel. A Gaussian here is a soft three-dimensional blob with a position, shape and opacity. Together the blobs form an object that can be rendered from another viewpoint. The output remains arranged like a two-dimensional image, allowing an ordinary image network to predict a three-dimensional representation." }, { "title": "The back of the object is still a prediction", "text": "A pixel usually shows a visible surface, but the network can place a predicted blob away from that surface to describe hidden structure. Training examples teach likely object shapes. This is useful inference, not a measurement of the unseen back. Two physical objects can look similar from the front while having different backs. An attractive new view does not uniquely identify which object was photographed." }, { "title": "Reconstruction and rendering are different speeds", "text": "First the network predicts the blobs; then a renderer draws views of them. In a classroom example, reconstructing an object in one second and rendering one hundred views in another second takes two seconds for the whole job. A hundred-frames-per-second rendering claim alone would omit the first stage. The paper likewise distinguishes encoding and rendering rates." }, { "title": "Object preparation and cameras remain part of the job", "text": "The source’s real-data experiments use object-centric preprocessing, foreground masks and reference scale or distance information. A multi-view extension needs relative camera poses. Some individual quality measures favor competitors, and training still consumes substantial compute. The method adds a fast prediction design while retaining the separate contribution of earlier Gaussian rendering. Before comparing two outputs, check whether both methods received the same mask and camera information. A cleaner input can improve a result without showing that the reconstruction algorithm alone is better." }, { "title": "Test the contract on a paper sketch", "text": "Draw a front view of a box, then sketch two different possible backs. Ask your agent to explain why the input cannot distinguish them without more evidence. Next, make a source table separating preprocessing, camera information, prediction time and rendering time. Someone else can continue by checking a missing condition rather than repeating an unqualified speed claim." } ], "references": [ { "title": "Splatter Image: Ultra-Fast Single-View 3D Reconstruction", "url": "https://arxiv.org/abs/2312.13150v1", "locator": "Original complete main method, object preprocessing, timing and training sections, comparison tables and conclusion." } ] }, "reading_status": "EXPLANATORY_GUIDE" }