Revisiting 3D LLM Benchmarks: Are We Really Testing 3D Capabilities?
This addresses a critical evaluation flaw for researchers and developers working on 3D LLMs, though it is incremental as it focuses on improving existing benchmarks rather than introducing new methods.
The paper identifies the '2D-Cheating' problem in 3D LLM evaluation, where tasks can be solved by VLMs using rendered images, and proposes principles for better assessing genuine 3D understanding.
In this work, we identify the "2D-Cheating" problem in 3D LLM evaluation, where these tasks might be easily solved by VLMs with rendered images of point clouds, exposing ineffective evaluation of 3D LLMs' unique 3D capabilities. We test VLM performance across multiple 3D LLM benchmarks and, using this as a reference, propose principles for better assessing genuine 3D understanding. We also advocate explicitly separating 3D abilities from 1D or 2D aspects when evaluating 3D LLMs. Code and data are available at https://github.com/LLM-class-group/Revisiting-3D-LLM-Benchmarks