Enhancing Instance-Level Image Classification with Set-Level Labels
Authors: Renyu Zhang, Aly A Khan, Yuxin Chen, Robert L. Grossman
ICLR 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conducted experiments on two distinct categories of datasets: natural image datasets and histopathology image datasets. Our experimental results demonstrate the effectiveness of our approach, showcasing improved classification performance compared to traditional single-instance label-based methods. |
| Researcher Affiliation | Academia | Renyu Zhang Department of Computer Science University of Chicago zhangr@uchicago.edu Aly A. Khan Department of Pathology and Family Medicine University of Chicago aakhan@uchicago.edu Yuxin Chen Department of Computer Science University of Chicago chenyuxin@uchicago.edu Robert L. Grossman Department of Computer Science and Medicine University of Chicago rgrossman1@uchicago.edu |
| Pseudocode | Yes | The pseudocode for FACILE is provided in Algorithm 1, and we further illustrate the FACILE algorithm in figure 3. |
| Open Source Code | No | The paper does not provide an explicit link or statement about the open-source availability of its code. Footnote 1 states 'We will make data publicly available upon acceptance of our paper', which refers to data and is a future promise, not current code availability. |
| Open Datasets | Yes | For natural image datasets, we sample input sets from training data from CIFAR-100... We pretrain our models using two independent sources of WSIs. First, we downloaded data from The Cancer Genome Atlas (TCGA) from the NCI Genomic Data Commons (GDC) (Heath et al., 2021)... Second, we downloaded all clinical slides from the Genotype-Tissue Expression (GTEx) project (Lonsdale et al., 2013) |
| Dataset Splits | No | The paper describes 'support set' (training) and 'query set' (testing) for its few-shot learning setup, but does not explicitly mention or detail a separate 'validation' split for hyperparameter tuning or model selection. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., GPU models, CPU types, memory specifications) used for running the experiments. |
| Software Dependencies | No | The paper mentions various software components and models (e.g., ResNet18, ViT-B/16, CLIP, DINO V2, SGD, faiss) but does not provide specific version numbers for these dependencies. |
| Experiment Setup | Yes | A.1 PRETRAIN WITH UNIQUE CLASS NUMBER AND MOST FREQUENT CLASS OF INPUT SETS: Sim Siam is trained for 2,000 epochs using a batch size of 512. SGD is employed with a learning rate of 0.1, weight decay of 1e-4, and momentum of 0.9. The training process incorporates a cosine scheduler. Similarly, Sim CLR is trained for 2,000 epochs with a batch size of 256 and a temperature of 0.07. SGD is used with a learning rate of 0.05, weight decay of 1e-4, and momentum of 0.9. The training also utilizes a cosine scheduler. |