Explain Any Concept: Segment Anything Meets Concept-Based Explanation

Authors: Ao Sun, Pingchuan Ma, Yuanyuan Yuan, Shuai Wang

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our evaluation over two popular datasets (Image Net and COCO) illustrate the highly encouraging performance of EAC over commonly-used XAI methods.
Researcher Affiliation Academia Ao Sun, Pingchuan Ma, Yuanyuan Yuan, and Shuai Wang The Hong Kong University of Science and Technology {asunac, pmaab, yyuanaq, shuaiw}@cse.ust.hk
Pseudocode No The paper includes a technical pipeline diagram (Figure 1) but does not provide any structured pseudocode or algorithm blocks.
Open Source Code Yes Open Source. We publicly release and maintain EAC under the following github page: https://github.com/Jerry00917/samshap.
Open Datasets Yes We evaluate EAC on two popular datasets, Image Net [38] and COCO [39]
Dataset Splits Yes We use the standard training/validation split for both datasets.
Hardware Specification No The paper does not provide specific hardware details such as GPU/CPU models, memory, or cloud instance types used for experiments.
Software Dependencies No The paper does not specify version numbers for key software components or libraries used in the experiments.
Experiment Setup Yes The only hyper-parameter considered in EAC is when fitting the PIE Scheme, i.e. a simple linear neural network learning scheme and the Monte Carlo (MC) sampling: we set lr = 0.008 and the number of MC sampling as 50000 throughout all experiments.