Daniel Wiesmann
Daniel Wiesmann
This is harder because the data is not in geotiff format. We have to evaluate how long it will take to do this first.
This is a proposal for a streaming approach that relies on static STAC Indexes and dynamic tiling. ### tl;dr - List imagery "scenes" using STAC items - Compute how many...
This model is currently training, this ticket can be completed once we are happy with the convergence. - [ ] Publish the latest weights on huggingface - [ ] Make...
Still draft, but works locally. I'll give this a go for 10 tiles so that @srmsoumya can test the data loaders with this libarary Refs #165
Running the trainer should work out of the box without requiring a wandb account. Refs: https://github.com/Clay-foundation/model/pull/116#discussion_r1451798021 https://github.com/Clay-foundation/model/pull/109#discussion_r1444170134
Ideas to add are - Train a larger model version - Use SAM as teacher model - Add Satellogic data to reduce bias on high res training
Added Modis to our training data set using stacchip. - Chip dates and locations data added to general index using the new [modis processor](https://github.com/Clay-foundation/stacchip/blob/main/stacchip/processors/modis_processor.py) in stacchip - All data prechipped...
Show how to use the clay torchscript encoder to create embeddings over AOI.
Small improvements on the naip sim search notebook in #247
We could then use HF links to open these instead of relative paths within the model repo, as we do in most scripts right now.