Internal Planning #1
Description
1. Proposed Logistics: (10 min)
1.1 Internal weekly meetings:
Each task is allocated 5 minutes for updates, and the task lead will present slides.
Upload the weekly slides into the shared GoogleDrive folder
1.2 Sync-up on Slack:
Every Tuesday morning, each task lead will do a quick sync-up on the ISI CriticalMAAS-AIM channel with 1-2 sentences. Describe what you'll be working on and what (if) you need from others.
1.3 External meetings with other TA1 (and TA4) performers:
[TO_BE_SCHEDULED] First, find a time that works for most of us (USC/UMN/Inferlink). Ideally, we'll have one person from each org to attend the meetings and then sync-up in our internal meeting.
All internal and external meetings will be open with a dedicated shared calendar so that DARPA/USGS/other performers can join if they like.
2. Shared Repos & Google Drive Folders (5 min)
- Github Project Repo: https://github.com/knowledge-computing/CriticalMAAS_AIM
- Send Zekun your GitHub ID to grant access to the repo
- Google Drive: https://drive.google.com/drive/folders/1SPdN8hxFWTZRwf4MAykvZFgwWmY3bcQk
- Project Website: https://github.com/knowledge-computing/criticalmaas-aim
3. Discuss about the proposed tasks (Summary Doc) (30min)
3.1. Leads (10 min)
3.1.1. Line Feature Extraction @weiwei
- Input: Scanned raster map & cropped line symbol
- Output: Binary segmentation mask (.geotif), vector with attributes (.geojson, decide on the schema, e.g., type, map URI, processing model)
- To discuss🧐 : output data schema
3.1.2 Text spotting on the map images (@leeje Jang)
- Input: Scanned raster map
- Output: Geojson that contains spotted labels (follow mapKurator schema)
3.1.3 Build overall end-to-end system and data production (@inferlink)
- Needs to collaborate with UIUC and Uncharted to incorporate other modules
- Input: Scanned raster map
- Output: Outputs from all the modules
- To discuss: 🧐
Define the required modules and input/output of each module
Whether to use mapKurator?
3.2 Other Components (20 min)
3.2.1 Point Feature Extraction @ju Sun @leeje
- Input: Scanned raster map & cropped point symbol
- Output: Binary segmentation mask (.geotif), vector with attributes (shapefile, decide on the schema, e.g., type, map URI, processing model)
- To discuss::monocle_face: output data schema
3.2.2 Polygon Feature Extraction & cropped polygon symbol @Fandel
- Input: Scanned raster map
- Output: Binary segmentation mask (.geotif), vector with attributes (shapefile, decide on the schema, e.g., type, map URI, processing model); List of bounding boxes for symbol detection(needs paired bounding boxes for symbol and text description)
- To discuss: 🧐 output data schema
3.2.3 Map area segmentation (New task) @Fandel
- Input: Scanned raster map
- Output: Binary segmentation mask (.tif/.png/.jpg) that has map area segmented
3.2.4 Georeferencing (New task) @zekun @Theresa
- Input: Scanned raster map
- Output: lat & lng of map corners?
- To discuss::monocle_face: Output data schema
4. Goals & Milestones (5 min)
-
Month 3: Internal prototype end-to-end system ready
Integrated with competition models- DARPA's Expected Outcomes (for the hackathon)
• Establish working relationships among performers within and across TAs
• USGS SMEs provide details and/or clarifications of CMA processes
• Reach agreement on data and file formats, naming conventions, etc.
• Develop roadmap for meeting Month 6 Baselining Event objectives
▪ Not expecting a usable prototype system that integrates all TAs - No formal evaluation, but...
• The Government team will assess progress towards program objectives
• Performers will be provided with feedback, and guidance as needed
- DARPA's Expected Outcomes (for the hackathon)
-
Month 6: TA1 end-to-end system ready for MVT Zinc
Integrate with other performers- DARPA's Expected Outcomes (for the hackathon)
Baselining Event at 6 Months
• Each TA (1-3) will be evaluated separately.
• Expert users will use tools to perform TA-specific tasks.
• Output will be assessed for quality (and throughput) and will receive qualitative feedback.
• This is an opportunity to debug each workflow, find out what works and what does not (can the
user accomplish what is needed). - EXAMPLE: MVT Zinc
https://drive.google.com/file/d/1LiwnMo8-ShJsZmtIpHr9TrTuAiwek82K/view?usp=sharing
- DARPA's Expected Outcomes (for the hackathon)
5. Kick-off Data: (5 min)
~140 GB Competition data + NGMDB data (Access Link)
To discuss: 🧐 How to share? Where to store?