vinkmr / asicd Goto Github PK

Docstring and explanations for majority of the functions are missing.

First of all, I apologize, but I didn't know a better way to establish contact with you.

I'm studying regenerative energy systems at tu dresden and am currently working on a computer-controlled camera network. i would love to come back to your code. i have a few questions - do you think it would be possible to exchange ideas via skype? that would be outstanding.

Best regards
Paul Matteschk

There is no main.py landing file when switching into project directory.

Some automated tools might not work.
Difficult for initial users to kick-off project development.

We all love Tensorboard, now let's implement it.

• Apply Sun remover (ResNets have better cloud detection, but poor sun removing capability)
• Possibility of static postprocessing layer within the model, so that the loss is calculated with respect to the threshold output.

Now, we are using a hacky way of loading models, by actually defining the configuration in python and loading it into the training script. We have since learned that this is not the best practice. In the later stages, we will be migrating to .yaml file for model configuration under a models/ folder.

Downloading files from FTP server throws missing folder error when the dataset folder is not created manually.

Investigate how to remove the influence of the sun. Now it is applied in the preprocessing step, which allows VGG based model to perform better than ResNet, in terms of sun exclusiton. But ResNets have better cloud detection.