Cross-Scale MAE: A Tale of Multiscale Exploitation in Remote Sensing
Authors: Maofeng Tang, Andrei Cozma, Konstantinos Georgiou, Hairong Qi
NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental evaluations demonstrate that Cross-Scale MAE exhibits superior performance compared to standard MAE and other state-of-the-art remote sensing MAE methods. (...) 4 Experiments & Results |
| Researcher Affiliation | Academia | 1Min H. Kao Department of Electrical Engineering and Computer Science University of Tennessee, Knoxville {mtang4, acozma, kgeorgio}@vols.utk.edu, hqi@utk.edu, |
| Pseudocode | No | The paper describes the model architecture and training process in text and diagrams (Fig. 1) but does not provide pseudocode or algorithm blocks. |
| Open Source Code | Yes | The details of the implementation and ablation studies on attention and loss types have been placed in the supplementary. |
| Open Datasets | Yes | We pre-train Cross-Scale MAE with a Vi T-Large model (unless mentioned otherwise) using the Functional Map of the World (f Mo W) [11] RGB training set, which consists of 363.6k images of varying image resolution and GSD. The datasets we use include RESISC45 [10], WHU-RS19 [47], UC Merced [47], Euro SAT [22], as shown in Table 1. |
| Dataset Splits | No | The paper mentions using 'f Mo W-RGB training set' for pre-training and then testing on various datasets for downstream tasks and KNN performance, but it does not explicitly provide details on training, validation, and test dataset splits (e.g., percentages or counts) for all experiments, nor does it explicitly mention a dedicated validation set with specific proportions. |
| Hardware Specification | Yes | Thus, for this comparison, we stress-test the x Formers and Timm versions of the baseline implementations by training them on individual Nvidia RTX A6000s. |
| Software Dependencies | No | The paper mentions using the 'x Formers library' but does not specify its version number or the versions of any other key software dependencies. |
| Experiment Setup | Yes | We fine-tune the model for 50 epochs for all downstream tasks, following the same hyper-parameter settings as Scale-MAE [36]. (...) This experiment is conducted on RESISC45 at two scaling ratios, 0.5, 1, on the raw image with Vi T-Base and pre-trained 300 epochs on f Mo W-RGB dataset with input size of 128 128. |