d-diaz / stand_mapping Goto Github PK

PyTorch's default Mask-RCNN implementation expects input formatted like this:

During training, the model expects both the input tensors, as well as a targets (list of dictionary), containing:

• boxes (FloatTensor[N, 4]): the ground-truth boxes in [x1, y1, x2, y2] format, with values of x between 0 and W and values of y between 0 and H
• labels (Int64Tensor[N]): the class label for each ground-truth box
• masks (UInt8Tensor[N, H, W]): the segmentation binary masks for each instance

We will probably want to implement a DataSet class to make data loading and any transformations straightforward. Here's a tutorial in PyTorch docs on making a DataSet. Here is an example specific to creating a DataSet for Mask-RCNN .

For HW3, you developed a few initial functions for your project. Can you include them somewhere in your repository, even if they're still rough? They don't need to be on the master branch either, but I couldn't find them on any other branches or forks of the project.

Issue for HW2-Q3
We haven't had a discussion yet so creating this placeholder issue in order to discuss the next steps
-aj-22 (Ajinkya Sheth)

Save to ./data folder on google drive

Utility functions that, given a bbox with extent in UTM coordinates:

• fetch landcover classification image from Microsoft AI for Earth landcover API
• fetch NAIP 4-band image
• fetch aboveground biomass 1-band image

You have included a short description of the scope of your project to your README.md file. Can you also add in, whether in the README or in your documentation, a list of use cases, or some other way of describing the specific instances your code can be applied? If stumble upon your repo, it woul be nice to quickly tell the available functionality.

I need to learn skimage!

Produce a notebook that converts WA DNR RSFRIS geodatabase into a shapefile that includes a consolidated set of land cover types

You probably have a rough idea of the software packages you intend to use to interact with your data and do some analyses. Somewhere in your documentation, can you include a list of dependencies?

I'm excited that you're using Sphinx for your project documentation. I know you're near the beginning of your project, but I'd like you to get a test-run of your docs going to make sure Sphinx is working correctly. If you can also provide a link somewhere, say in your README.md file, so readers can quickly get to your documentation, that would be great.

Create a document that elaborates on the primary data acquisition, cleaning, and modeling stages for this project.

First, I think I will go through the material we have.
Second, we will have more discussion and to see the job distribution.

Add a script, scoring.py to ./stand_mapping/utils/. Create functions which will enable the scoring scoring of a single-channel integer-format image with a proposed segmentation against a ground-truth segmentation. Should return a score indicating the average IOU across all segments matched to ground-truth segments. This functions from the matterport implementation of Mask-RCNN look like a good place to start.

If we want to use these matterport functions, we will need to implement an additional function that converts a single-channel proposed segmentation (array of shape (W, H) or (W, H, 1)) into an array of segment masks with shape (W, H, num_segments) where each channel in this array contains the mask for an individual segment.

We may also need to filter the segments used for this score to focus only on forested areas (ignoring segments that are dominated by water, field, or built-up land cover types).