RGMIL: Guide Your Multiple-Instance Learning Model with Regressor
Authors: Zhaolong Du, Shasha Mao, Yimeng Zhang, Shuiping Gou, Licheng Jiao, Lin Xiong
NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In experiments, RGP shows dominance on more than 20 MIL benchmark datasets, with the average bag-level classification accuracy close to 1. We also perform a series of comprehensive experiments on the MMNIST dataset. Experimental results illustrate that our aggregator outperforms existing methods under different challenging circumstances. Instance-level predictions are even possible under the guidance of RGP information table in a long sequence. |
| Researcher Affiliation | Collaboration | Zhaolong Du1, Shasha Mao1 , Yimeng Zhang1, Shuiping Gou1, Licheng Jiao1, Lin Xiong2 1Xidian University, 2Sense Time. |
| Pseudocode | No | The paper describes the proposed method using mathematical equations and text, but it does not include a formal pseudocode block or algorithm. |
| Open Source Code | Yes | The codes are available on https://github.com/LMBDA-design/RGMIL. |
| Open Datasets | Yes | With the branch number N = 1, five classic MIL benchmark datasets [3][1] are used to evaluate the bag-level performance... Thus, we constructed a flexible MIL dataset based on MNIST [14] and denoted it as MMNIST... we use the benchmark pain dataset (UNBC-Mc Master Shoulder Pain dataset [17], UNBC)... To validate the performance of RGMIL on the general multi-class bag-level classification problems, we test our model on SIVAL [20] dataset... |
| Dataset Splits | Yes | The results are the average of five independent experiments with a 10-fold cross-validation... To compare with the existing supervised models, we follow the same experimental settings with the S-O-T-A model MSRAN (ICBBT 21) [2]: using 25-fold cross-validation with a leave-one-subject-out strategy and evaluating by four metrics. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., CPU, GPU models, or memory specifications) used for running the experiments. |
| Software Dependencies | No | The paper does not specify version numbers for any software dependencies or libraries used in the implementation. |
| Experiment Setup | Yes | In the branch Bi, given the instance-level representation matrix fs Rc t, the RGP works as follows: H = fs ; Pk = W i Hk + bi; (k [1, t], P Rt 2) ... For the test, we use a fixed number of 10000 test images... all images are resized as (3, 224, 224), and the training bag size is set as 64. An average of about 6000 bags is used as training data during each fold validation, and Res Net18 [8] is used as backbone. In this experiment, we take consecutive 64 frames in a video as a training bag. To increase the amount of data, we used a sliding window approach with a step size of 8 to produce training bags. |