SLiMe: Segment Like Me

Authors: Aliasghar Khani, Saeid Asgari, Aditya Sanghi, Ali Mahdavi Amiri, Ghassan Hamarneh

ICLR 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Through broad experiments, we examined various design factors and showed that SLi Me outperforms existing oneand few-shot segmentation methods.
Researcher Affiliation Collaboration Aliasghar Khani1,2, Saeid Asgari Taghanaki1,2, Aditya Sanghi1, Ali Mahdavi Amiri2, Ghassan Hamarneh2 1 Autodesk Research 2 School of Computing Science, Simon Fraser University
Pseudocode No The paper describes the methodology in prose and through diagrams (e.g., Figure 3 and 4), but does not provide any explicitly labeled pseudocode or algorithm blocks.
Open Source Code Yes The source code of the project is publicly available.
Open Datasets Yes We use m Io U to compare our approach against three existing methods: Re GAN (Tritrong et al., 2021), Seg DDPM (Baranchuk et al., 2021), and Seg GPT (Wang et al., 2023) on two datasets: PASCALPart (Chen et al., 2014) and Celeb AMask-HQ (Lee et al., 2020).
Dataset Splits Yes We present the results for both 10-sample and 1-sample settings, utilizing a single validation sample for 10-sample experiments of SLi Me.
Hardware Specification No The paper does not specify any hardware details such as GPU models, CPU types, or memory used for running the experiments.
Software Dependencies Yes We opted for SD version 2.1
Experiment Setup Yes For optimization, we employed the Adam optimizer with a learning rate of 0.1, optimizing our method for 200 epochs with a batch size of 1. Additionally, we used weighted cross-entropy loss, with each class s weight determined as the ratio of the number of whole pixels in the image to the number of pixels belonging to that class within the image. Furthermore, the values for α and β were set to 1 and 0.005, respectively. We set H and W to be 64. For optimizing on PASCAL-Part classes, we applied the following augmentations: Random Horizontal Flip, Gaussian Blur, Random Crop, and Random Rotation. For the car class, we set the random crop ratio range to [0.5, 1], while for the horse class, it was adjusted to [0.8, 1]. Additionally, we applied random rotation within the range of [ 30, 30] degrees.