Closed Assylnur closed 4 months ago
Hi, thank you for your interest in the project!
1) Rather than keeping the black images in your timeseries, I would recommend repeating the good images. Our pretrained models don't see many (if any) black images, so I don't think it would work well. 2) Our models are trained exclusively on L1C imagery. Regarding L2A imagery, I have not tried inputting L2A into our pretrained models. I imagine it wouldn't work super well, because of the very different preprocessing, but you could try. If you are able to, you could try fine-tuning the model on L2A imagery.
Hi,
Thank you for your informative response! I have a few follow-up questions to better understand how to proceed:
1) Could you advise on where I can find L1C imagery?
2) I attempted to use L2A data with your pretrained models but, as you said, the results were not optimal.
input:
Result:
Could you provide any recommendations for the fine-tuning the model with L2A imagery?
Thank you once again for your support.
1) If you want to download Sentinel-2 L1C imagery from the source, see here:. We have details on how to process raw imagery into the format our models expect here.
2) To finetune our model on L2A imagery, you might try loading in pretrained weights (from here) and then trying to train on your data. I'm not sure how well it will work, it will depend on how much imagery you have.
I tried using Sentinel-2 L1C data as input but I'm still encountering the same issues with the results. Could you please review my inputs? my input data: input.zip
Thank you for the resources and recommendations
Sorry, I missed this! Could you send the coordinates of this tile, and the super-res outputs that you're getting with this input?
Hello,
I only have access to Sentinel-l2A data for my project and am interested in applying your model. So, can I use Sentinel-2A data as input for the model?
Some of the images in my timeseries image are black:
Will they impact the overall performance or require any special preprocessing?
Your insights on these issues would be greatly appreciated. Thank you for your time and assistance.