SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2402.07865 · ICML · 2024

Prismatic VLMs: Investigating the Design Space of Visually-Conditioned Language Models

J Kim, Percy Liang, Siddharth Karamcheti, Dorsa Sadigh, Ashwin Balakrishna, Suraj Nair, C Kim, M Lin, S Lin, P Lue, A Krueger, O Keefe, and 2 more

arXiv · PDF · Open in the Atlas

Code that ran

We lifted 5 functions out of this paper's own repositories and ran 1 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.

RepositoryRoleRan
tri-ml/prismatic-vlms canonical 1 of 1
tri-ml/vlm-evaluation canonical 0 of 4
FunctionStatusWhere it lives
initialize_overwatch Ran tri-ml/prismatic-vlms/prismatic/overwatch/overwatch.py
code served (permissive licence) · get_code("6e9a835262e7ec57")
ImageProcessor Not yet run tri-ml/vlm-evaluation/vlm_eval/models/prismatic.py
pointer only (licence: NOASSERTION) · get_code("08e6aeaa327cb1f2")
PrismaticVLM Not yet run tri-ml/vlm-evaluation/vlm_eval/models/prismatic.py
pointer only (licence: NOASSERTION) · get_code("1c76ba920eb87f9b")
Tokenizer Not yet run tri-ml/vlm-evaluation/vlm_eval/models/prismatic.py
pointer only (licence: NOASSERTION) · get_code("e0633b98873dbfc7")
VLM Not yet run tri-ml/vlm-evaluation/vlm_eval/models/prismatic.py
pointer only (licence: NOASSERTION) · get_code("a2488e8a8252030f")

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Visually-conditioned language models (VLMs) have seen growing adoption in applications such as visual dialogue, scene understanding, and robotic task planning; adoption that has fueled a wealth of new models such as LLaVa, Instruct-BLIP, and PaLI-3. Despite the volume of new releases, key design decisions around image preprocessing, architecture, and optimization are underexplored, making it challenging to understand what factors account for model performance -a challenge further complicated by the lack of objective, consistent evaluations. To address these gaps, we first compile a suite of standardized evaluations spanning visual question answering, object localization, and challenge sets that probe properties such as hallucination; evaluations that provide fine-grained insight VLM capabilities. Second, we rigorously investigate VLMs along key design axes, including pretrained visual representations and training from base vs. instruct-tuned language models, amongst others. We couple our analysis with three resource contributions: (1) a unified framework for evaluating VLMs, (2) optimized, flexible training code, and (3) checkpoints for all models, including a family of VLMs at the 7-13B scale that strictly outperform InstructBLIP and LLaVa v1.5, the state-of-the-art in open VLMs. * Prismatic (adj) -relating to or having the form of a prism. Like a geometric prism, our VLMs share a common structure, but are characterized by different "faces" -the individual design axes we explore in this work.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2402.07865")
get_code_for_paper("2402.07865")
have("2402.07865")

Connect an agent — have() is free.