dsen2-cr
dsen2-cr copied to clipboard
ameraner
Prediction model returns the inputed photo with clouds
After installing said libraries and formatting csv file as it's described in the readme file I managed to make the code run, however, the predicted output is exactly like the photo with cloud I have inputted. Has anyone else encountered something similar? What should I do to get the actual result? Thank you
Hi @RMNT , thank you for your feedback. This is strange. One thing I can think of is that somehow you're training using the cloudy images as target. Are you sure that your csv setup/column order is correct?
Test with this environment instructions on Anaconda, working for me
conda create -y -n dsen2_cr_v1 python=3.7.0 conda activate dsen2_cr_v1 conda install numpy=1.17.5 pip install tensorflow==1.13.2 pip install keras==2.2.4 pip install matplotlib==2.1.1 pip install rasterio pip install pydot pip install h5py pip install graphviz pip install scipy pip install protobuf==3.6.1
Hi @RMNT , thank you for your feedback. This is strange. One thing I can think of is that somehow you're training using the cloudy images as target. Are you sure that your csv setup/column order is correct?
Thank you for your reply. I am using provided weights (full DSen2-CR model trained by optimizing the CARL loss) and am only testing the model. I double checked the csv file column order and it is [sample type, sentinel-1, sentinel-2 cloud free, sentinel-2 with clouds and the final one is file name]. Example: 3,s1/,noclouds/,clouds/,p14617.tif
Hi @ameraner , I am also not able to see the output as expected. There is also some cloud on the prediction. Could you please let me know how many epochs you ran and what is your datasize? I just want to whether your model trained properly or not?
Thank you
Hi @StrangeReseach100PK , all information on the dataset and the training setup is contained in the paper (it is open-access here). Please let me know if you need any more information than what is described there.
Closing this as the questions have been answered, feel free to reopen if you have any more issues/questions!