Visual Question Answering On Clevr

Metrics

Accuracy

Results

Performance results of various models on this benchmark

		Paper Title
NS-VQA (1K programs)	99.8	Neural-Symbolic VQA: Disentangling Reasoning from Vision and Language Understanding
MDETR	99.7	MDETR -- Modulated Detection for End-to-End Multi-Modal Understanding
NeSyCoCo	99.7	NeSyCoCo: A Neuro-Symbolic Concept Composer for Compositional Generalization
OCCAM (ours)	99.4	Interpretable Visual Reasoning via Induced Symbolic Space
TbD + reg + hres	99.1	Transparency by Design: Closing the Gap Between Performance and Interpretability in Visual Reasoning
MAC	98.9	Compositional Attention Networks for Machine Reasoning
NS-CL	98.9	The Neuro-Symbolic Concept Learner: Interpreting Scenes, Words, and Sentences From Natural Supervision
CNN + LSTM + RN + HAN	98.8	Learning Visual Question Answering by Bootstrapping Hard Attention
DDRprog*	98.3	DDRprog: A CLEVR Differentiable Dynamic Reasoning Programmer
single-hop + LCGN (ours)	97.9	Language-Conditioned Graph Networks for Relational Reasoning
XNM-Det supervised	97.7	Explainable and Explicit Visual Reasoning over Scene Graphs
CNN+GRU+FiLM	97.7	FiLM: Visual Reasoning with a General Conditioning Layer
IEP-700K	96.9	Inferring and Executing Programs for Visual Reasoning
CNN + LSTM + RN	95.50	A simple neural network module for relational reasoning
QGHC+Att+Concat	65.90	Question-Guided Hybrid Convolution for Visual Question Answering

0 of 15 row(s) selected.