Robin Cole
Robin Cole
I've now created a couple of streaming datasets and have a common class that can handle them all - it just requires the datasets are formatted in a standard way...
Possibly [Spacenet 7](https://spacenet.ai/sn7-challenge/) - but broadly speaking change detection is a special case of time series, so an API that extends `image1, image2, 3,4 etc` makes sense
Reading the [paper](https://arxiv.org/pdf/1805.06561.pdf), appears the imagery is from deepglobe/maxar, will check my network for contacts there
RE license, I came to a dead end - the people I spoke to who had worked on the dataset have all moved on As to updating the docs, I...
I think at a minimum we should improve the documentation to state when augmentations are and are not applied. For example I assumed they are performed in the dataset get_item,...
Thanks for the clarification - it makes sense that there are basic augmentations to always apply during training and which we don't need to inspect (ie normalisation) and others to...
The adaption for multi-channel looks straightforward - not sure about handling the pretrained weights https://github.com/allenai/vessel-detection-sentinels/blob/main/src/models/frcnn.py
In my own plotting funcs I have been applying the `percentile_normalisation` to 'undo' the effects of normalisation, but the result is not always that great. I think an `un-normalize` followed...
Thanks for the heads up, I will update the notebook example when I get time
`//10.20.10.106/config/` does not look correct