An Empirical Study of GPT-3 for Few-Shot Knowledge-Based VQA

By Zhengyuan Yang, Zhe Gan, Jianfeng Wang, Xiaowei Hu, Yumao Lu, Zicheng Liu, Lijuan Wang. 2021

537 citations recorded by the source corpus.

Knowledge-based visual question answering (VQA) involves answering questions that require external knowledge not present in the image. Existing methods first retrieve knowledge from external resources, then reason over the selected knowledge, the input image, and question for answer prediction. However, this two-step approach could lead to mismatches that potentially limit the VQA performance. For example, the retrieved knowledge might be noisy and irrelevant to the question, and the re-embedded knowledge features during reasoning might deviate from their original meanings in the knowledge base (KB). To address this challenge, we propose PICa, a simple yet effective method that Prompts GPT3 via the use of Image Captions, for knowledge-based VQA. Inspired by GPT-3’s power in knowledge retrieval and question answering, instead of using structured KBs as in previous work, we treat GPT-3 as an implicit and unstructured KB that can jointly acquire and process relevant knowledge. Specifically, we first convert the image into captions (or tags) that GPT-3 can understand, then adapt GPT-3 to solve the VQA task in a few-shot manner by just providing a few in-context VQA examples. We further boost performance by carefully investigating: (i) what text formats best describe the image content, and (ii) how in-context examples can be better selected and used. PICa unlocks the first use of GPT-3 for multimodal tasks. By using only 16 examples, PICa surpasses the supervised state of the art by an absolute +8.6 points on the OK-VQA dataset. We also benchmark PICa on VQAv2, where PICa also shows a decent few-shot performance.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Multimodal Learning

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Multimodal Learning

Citations in the corpus

Socratic Models: Composing Zero-Shot Multimodal Reasoning with Language (2022, background): “As in [81] we can also further prompt the LM to explain the answer by adding “This is because:” to which it can respond “I saw on the porch a package and knew that I was expecting it.”
Flamingo: a Visual Language Model for Few-Shot Learning (2022, background): “One recent line of work [26, 68, 78, 114, 136, 144] proposes to freeze the pretrained LM weights to prevent catastrophic forgetting [71].”
KAT: A Knowledge Augmented Transformer for Vision-and-Language (2021, methodology): “Further, external APIs, such as Google (Wu et al., 2021; Luo et al., 2021), Microsoft (Yang et al., 2021), and OCR (Luo et al., 2021; Wu et al., 2021) are used to enrich the associated knowledge.”
VL-ADAPTER: Parameter-Efficient Transfer Learning for Vision-and-Language Tasks (2021, methodology): “Although language models are tuned for downstream tasks in previous works, some recent research [49, 51] attempts to freeze large language models (e.g., GPT-3) to achieve zero-shot learning for V&L tasks.”
MAGMA - Multimodal Augmentation of Generative Models through Adapter-based Finetuning (2021, methodology): “SOTA model SimVLM [55] PICa [58] CFR [33] Pythia [46] SimVLM SimVLM VIVO [22] SimVLM OSCAR [29]”
ViperGPT: Visual Inference via Python Execution for Reasoning (2023, neutral): “TRiG [13] 50.5 KAT [16] 54.4 RA-VQA [32] 54.5 REVIVE [33] 58.0 PromptCap [21] 58.8 Z S PNP-VQA [52] 35.9 PICa [60] 43.3 BLIP-2 [30] 45.9 Flamingo [1] 50.6 ViperGPT (ours) 51.9”
Visual ChatGPT: Talking, Drawing and Editing with Visual Foundation Models (2023, background): “With the increased number of model parameters, it is hard to train those pretrained LLMs, thus more efforts have been paid to directly leverage the off-the-shelf frozen pre-trained LLMs for VL tasks [12, 38, 8, 46, 50].”
GIT: A Generative Image-to-text Transformer for Vision and Language (2022, background): “Meanwhile, different test image requires different support shots based on Yang et al. (2022).”
Plug-and-Play VQA: Zero-shot VQA by Conjoining Large Pretrained Models with Zero Training (2022, methodology): “Most similar to our work, PICa (Yang et al., 2022) converts an image to a single caption and adopts GPT-3 (Brown et al., 2020) for zero-shot VQA.”
All You May Need for VQA are Image Captions (2022, background): “Yang et al. (2021b) prompt the GPT-3 (Brown et al., 2020) to answer knowledge-based visual questions based on generated captions and tags and a few VQA examples.”
OpenFlamingo: An Open-Source Framework for Training Large Autoregressive Vision-Language Models (2023, methodology): “Full results with RICES are listed in Table 12 and illustrated in Figure 10.”
BLINK: Multimodal Large Language Models Can See but Not Perceive (2024, result): “Such a conflation has also been reported for other benchmarks by previous work [11,38,84].”
Zero-Shot Video Question Answering via Frozen Bidirectional Language Models (2022, methodology): “Recently, a promising line of work builds on frozen large autoregressive language models [22, 75, 102, 107, 116, 123] for zero-shot visual question answering.”
GPT4Tools: Teaching Large Language Model to Use Tools via Self-instruction (2023, methodology): “Given an image, we construct the image content X C according to the captions and bounding boxes, which is a straightforward means of establishing connections between an image and a language model [17, 18].”