Model download folder: names truncated in Google Drive, making it impossible to identify models before downloading

[chuckhenrich]

The model download folder on Google Drive uses folder names that encode the full training command. These names are very long and truncated by Google Drive's UI, making it impossible to identify which model is which without downloading and extracting each zip first. (https://drive.google.com/drive/folders/1XmY9yO3yhhhdwQ_btYCIpkUBFQ88H-pr)

The zip files themselves don't need to be named with all training command parameters to make them identifiable to users. The folder names inside the zip files include those parameters.

So I propose two alternative approaches to make the downloads easier for people to identify and choose:

1. Rename the download files to shorter, user-friendly names, for example, rename the folders on Google Drive:

"2021-06-14T20_27_nn_train.py_--config_configs-train_conf_utnet_std.yaml_--config2_configs-train_with_clean_data.yaml_--g_model_path_..-..-models-nind_denoise-2021-06-12T11_48_nn_train.py_--config_configs-train_conf_utnet_std.yaml_--config2_configs-train_w.zip"

becomes

"2021-06-14-UtNet-650"

so the user can identify the core differentiators among the options.

2. Instead of option 1 (or even in addition) include a README in the Google Drive folder, mapping each zip file to a plain-English description (architecture, epoch count, notable training parameters) to make the models more understandable to users.

Something like this?

# nind-denoise models ReadMe

## Available models

| Zip file (shortened) | Architecture | Epochs | Notes 
|----------------------|--------------|--------|-------
| 2021-06-14T20:27_... | UtNet        | 650    | Trained with clean data augmentation — recommended
| 2021-05-31T22:11_... | UtNet        | 1000   | Hardswish activation 
| 2021-05-29T17:51_... | UtNet        | 5000   | Standard 
| 2021-05-23T10:16_... | UNet         | 1000   | Standard 
| 2019-08-03T16:14_... | UNet         | 280    | Earlier model 
| 2019-02-18T20:10_... | UNet         | 257    | Original 2019 research model 

## Which model should I use?

Start with the **2021-06-14 UtNet 650** model (trained with clean data). It generally produces the best results.

If that gives unsatisfactory results for your images, try the other UtNet models. Fall back to UNet models only if UtNet models don't work well for your use case.

Of course, you should make sure the details are correct.

I'm happy to contribute to documentation if you need any help.