Label super-resolution networks
Authors: Kolya Malkin, Caleb Robinson, Le Hou, Rachel Soobitsky, Jacob Czawlytko, Dimitris Samaras, Joel Saltz, Lucas Joppa, Nebojsa Jojic
ICLR 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We test our proposed algorithm in a challenging land cover mapping task to super-resolve labels at a 30m resolution to a separate set of labels at a 1m resolution. We compare our algorithm with models that are trained on high-resolution data and show that 1) we can achieve similar performance using only low-resolution data; and 2) we can achieve better performance when we incorporate a small amount of highresolution data in our training. We also test our approach on a medical imaging problem, resolving low-resolution probability maps into high-resolution segmentation of lymphocytes with accuracy equal to that of fully supervised models. |
| Researcher Affiliation | Collaboration | Kolya Malkin1,2 Caleb Robinson1,3 Le Hou1,4 Rachel Soobitsky5 Jacob Czawlytko5 Dimitris Samaras4 Joel Saltz4 Lucas Joppa1 Nebojsa Jojic1 1Microsoft Research 2Yale University 3Georgia Institute of Technology 4Stony Brook University 5Chesapeake Conservancy |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper mentions an interactive web application accessible online at http:// landcovermap.eastus.cloudapp.azure.com:4040/ that lets users query their models, but it does not state that the source code for their methodology is publicly available or provide a link to a code repository. |
| Open Datasets | Yes | We use three datasets: 4-channel high-resolution (1m) aerial imagery from the US Department of Agriculture, expensive high-resolution (1m) land cover data covering the Chesapeake Bay watershed in the north eastern United States (Chesapeake Bay Conservancy (2016; 2017)), and much more widely available low-resolution (30m) NLCD land cover data (see Fig. 1 for examples of the data, and Appendix B). To evaluate the segmentation performance, we use the lymphocyte classification dataset introduced in Hou et al. (2018). |
| Dataset Splits | Yes | We divide these datasets into four geographical regions: Maryland 2013 training region with high-resolution training labels, Maryland 2013 test region, Chesapeake 2013 test region, and Chesapeake 2014 test region. Because all baseline CNNs require supervised data, they are all evaluated using four-fold crossvalidation on the aforementioned dataset of 1786 image patches. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., GPU model, CPU type, memory) used for running the experiments. |
| Software Dependencies | No | The paper mentions the CNTK framework but does not provide specific version numbers for it or any other software dependencies. |
| Experiment Setup | Yes | The model cores and label counting modules were implemented in the CNTK (Seide & Agarwal (2016)) framework and trained using RMSProp (Hinton et al.) with an initial learning rate of 10 3 decaying to 10 6 by a factor of 10 per 6000 minibatches. Each minibatch contained 20 patches of 240 240 pixels each. As the network architecture, we use a U-Net (Ronneberger et al. (2015)) with 4 down-sampling and 4 up-sampling layers. After each down-sample or up-sample, we apply three convolutional layers with 32 to 64 filters. We apply batch normalization (Ioffe & Szegedy (2015)) before every activation function, except for the final (pre-softmax) output. |