Skills Engineering Mastering Related Work for Graphics Papers

Mastering Related Work for Graphics Papers

v20260724
siggraph-related-work
Provides a comprehensive guide on structuring the 'Related Work' section for top-tier computer graphics conferences like SIGGRAPH. It teaches authors how to position a new contribution based on 'capability delta' (what prior methods fail to do) rather than chronology, covering specialized technical lanes (rendering, geometry, AI for graphics) and managing complex academic attribution.
Get Skill
357 downloads
Overview

SIGGRAPH Related Work

At SIGGRAPH, related work is where you convince a domain-expert reviewer that your contribution is new capability, not a rediscovery. Graphics reviewers know the canon and the current state-of-the-art in your sub-area intimately; a missed or mischaracterized prior method is a fast reject. Position by capability delta — what prior methods cannot do that yours can — not by chronology. Anchor venue facts to resources/official-source-map.md.

Know the lanes

SIGGRAPH's scope spans distinct sub-communities, each with its own canon and its own reviewers. Place your paper in the right lane(s) and cover that lane's recent state-of-the-art:

Lane What the lane cares about Adjacent venues to check
Rendering / light transport Noise, bias, convergence, speed, physical accuracy EGSR, HPG, TOG
Geometry processing Robustness on real meshes, guarantees, generality SGP, TOG
Animation / character Naturalness, control, temporal coherence SCA, TOG
Physical simulation Stability, energy behavior, time-step, scale SCA, TOG
Imaging / computational photography Reconstruction quality, artifacts, hardware ICCP, TOG
Geometry/appearance capture, fabrication Fidelity to real objects, manufacturability TOG, EG
Learning for graphics (neural rendering, generative 3D) Quality, generality, controllability CVPR/ICCV overlap, TOG
Interaction / VR-AR-MR, HCI-for-graphics Latency, presence, usability CHI/UIST overlap, TOG

A paper often sits in two lanes (e.g., neural rendering = rendering + learning). Cover both; a reviewer from either lane will check that their canon is represented.

Capability-delta positioning

For each closest prior method, state the axis on which you differ and by how much:

  • Quality: "Prior method X leaves visible noise below N samples; ours converges at N/4."
  • Speed: "X runs offline; ours is real-time at the same quality."
  • Generality: "X assumes manifold input; ours handles the non-manifold meshes practitioners actually have."
  • Robustness: "X diverges on stiff configurations; ours remains stable."

Never leave the comparison at "we are related to X." Name the axis, and back it with a comparison in the Results (see siggraph-experiments). The strongest baseline must appear both here and in a head-to-head figure/table.

Attribution discipline (venue collisions are common)

Graphics ideas migrate across venues; attribute precisely:

  • SIGGRAPH vs SIGGRAPH Asia vs TOG-direct. All three publish in ACM TOG. Cite the specific issue; do not say "SIGGRAPH" for a SIGGRAPH Asia or a rolling TOG paper.
  • SIGGRAPH vs Eurographics (EG/CGF). Many canonical techniques debuted at Eurographics or in Computer Graphics Forum, not SIGGRAPH — check dblp before attributing a method to SIGGRAPH.
  • Specialized venues. EGSR (rendering), SGP (geometry), SCA (animation/simulation), HPG (high-performance graphics), I3D (interactive 3D), ICCP (computational photography) host founding papers routinely misattributed to SIGGRAPH.
  • The vision overlap. Neural rendering / generative-3D work has a large CVPR/ICCV literature; cite it, and be clear which contribution is the graphics advance versus the vision one.

Self-positioning and blinding

SIGGRAPH review has historically been single-blind (authors visible), so citing your own prior work in the natural voice is usually fine — but confirm the current cycle's policy (待核实 for 2026). If a cycle requires anonymized review, cite your prior work in the third person and avoid "our previous system X."

Anti-patterns

  • A chronological literature tour with no capability delta.
  • Omitting the single strongest baseline because it is inconvenient — the reviewer knows it exists.
  • Attributing an Eurographics/EGSR/SCA method to SIGGRAPH (or vice versa) without checking dblp.
  • Treating a CVPR neural-rendering line as if it did not exist because "this is a graphics venue."
  • Positioning against a strawman instead of the current state-of-the-art in the lane.

Output format

[Lanes] which sub-community/-ies; is each lane's SOTA covered? yes/no
[Strongest baseline] named here and compared head-to-head in Results? yes/no
[Deltas] <prior method -> axis (quality/speed/generality/robustness) -> magnitude>
[Attribution] SIGGRAPH vs SA vs TOG vs EG/EGSR/SGP/SCA checked on dblp? yes/no
[Blinding] cycle policy confirmed; self-cite voice correct? yes/no
[Fixes] <ordered>
Info
Category Engineering
Name siggraph-related-work
Version v20260724
Size 4.87KB
Updated At 2026-07-29
Language