DV-499 ⁃ Identify countries for 1-step model and gather data

[sync-by-unito[bot]]

Identify countries with less varied habitats (so qualify for 1-step classifier), then gather and prep data.

Countries/regions to start with (after South Africa):

• Zambia - Lauren, Blessing and Jeff
• Senegal - Marine, Robin and Paolo Strampelli (lion scientist)
• Rest of West Africa (combo of countries, including Benin, Ghana, Ivory Coast; but excluding Senegal) - Marine, Robin and Paolo Strampelli (lion scientist)
• Saudi Arabia

Countries to do after:

• U.S. (might need to be specific to West coast US and East coast US)
• Central America (I think Central America is similar enough in vegetation and species that the entire region can continue to use a Central America classifier, as exists today)
• Colombia
• Chile (might need a mountain-focused South American classifier and a non-mountain-focused South American classifier (haha as you can tell, I’m really clued-in to South America details))
• (Brazil)
• (Peru)
• (Argentina)
• Gabon
• India
• Thailand
• Malaysia
• Indonesia

• Central Asia (needs to be confirmed whether we have enough images from this region to split it up any more)

  • note- countries with brackets () around it indicate countries in which a machine classifier has been requested, but we may lack images in order to train a classifier - does not currently have a classifier

STEPS INVOLVED:

1. Reach out to local team members to get cleanest survey list – to ensure the data that we use to train the classifier is as clean as possible (explain to them the concept of garbage in = garbage out)
2. Once the cleanest survey list is received, investigate number of images per species to identify species within that might not have enough images to use when training the machine classifier. These may be able to be binned/grouped with other species in a more general category.
3. Lastly, send the list of all species contained within those cleanest surveys to local team members to decide which species categories, if any, need to be binned/grouped, using their local knowledge of their region.

┆Issue is synchronized with this Jira Task by Unito

[sync-by-unito[bot]]

➤ Shannon Dubay commented:

Melvin Ollewagen , Thabied Majal - I have update the task description for this to include the steps that need to be completed per country, and a list of the 4 countries to start with. I also started a rough list of allllllll the other countries/regions that we’ll still need to do after the initial 4, just to remember that there are many, many more to do after these. Currently the list is sitting around a total of at least 15 models.

[sync-by-unito[bot]]

➤ Shannon Dubay commented:

Note- I know Marine and Robin have already provided some of the first steps for West Africa to Thabied Majal

[sync-by-unito[bot]]

➤ Thabied Majal commented:

Tracking suggestions and notes in Notion:

https://www.notion.so/Classifier-Region-specific-notes-d976960e406b4179b9f0f1a9d99601dd?pvs=4 ( https://www.notion.so/Classifier-Region-specific-notes-d976960e406b4179b9f0f1a9d99601dd?pvs=4 )

[sync-by-unito[bot]]

➤ Shannon Dubay commented:

UPDATE Thabied Majal. On a work planning call for Senegal data processing, it was highlighted by Phil Henshel (Central and West Africa Regional Director) that the classifier for Senegal specifically needs to be prioritized, for completion before July. This is because they have a Senegal significant dataset that needs to be processed in August. Phil suggested that Senegal (Niokolo Koba site) have its own classifier, then we create another classifier for the rest of West Africa (southern West Africa), including sites from Benin (sites: W, Penjari), Ghana (site: Mole), and Ivory Coast (site: Comoe, Tai). I have updated the above description to highlight this.

In Summary, moving forward please focus on Zambia and Senegal as the next classifiers to be produced.

[sync-by-unito[bot]]

➤ Shannon Dubay commented:

FYI Melvin Ollewagen , please see my previous task. Melvin Ollewagen and Thabied Majal should we break this up into smaller tasks?

[sync-by-unito[bot]]

➤ Melvin Ollewagen commented:

Shannon Dubay Based on the steps involved im happy to keep it in 1 ticket for now, also nice just for metrics sake

[sync-by-unito[bot]]

➤ Thabied Majal commented:

Shannon Dubay since this jire task is just to identify and itemize everything we need, I think it would be better to break it up into separate tasks AFTER the Clara work is done, so we can better track progress for each individual country/item/model in case stakeholders want to check on progress related to their specific model. We can add them as sub-tasks to the Machine Classifier Overhaul epic

[sync-by-unito[bot]]

➤ Shannon Dubay commented:

Sounds good Melvin Ollewagen , Thabied Majal , thanks for the input