ecko-cli 1.6.0

CLI tool that easily converts a directory of images into a dataset for training generative ai models

ecko-cli

[!IMPORTANT] This tools has the option to now use the JoyCaption model for captioning images. You will need to have huggingface hub installed to download the model and the tokenizer. Also JoyCaption is a large model and will require a GPU with a healthy amount of memory. Demo on Huggingface // Huggingface Repo

[!IMPORTANT] This tool makes use of the flash-attention library, which has known to be problematic to install based on PyTorch > and CUDA versions. You may need to install the dependencies manually if you encounter issues. The way to install flash-attention is to clone > the repo and install the package with pip. This is the recommended way to install the package. You can also install the package with pip, but you will need to clone the repo first

git clone https://github.com/Dao-AILab/flash-attention.git
cd flash-attention
pip install flash_attn --no-build-isolation
pip install timm

Install via pipx

pipx inject ecko-cli flash_attn
pipx inject ecko-cli timm

Install via the CLI

ecko tools install-flash-attention

[!NOTE] You may notice a delay when first using the tool...normally this means that the models are being downloaded/update from huggingface or that they are being moved to your GPU. Check your terminal to make see progress

Overview

ecko-cli is a simple CLI tool that streamlines the process of processingimages in a directory, generating captions, and saving them as text files. Additionally, it provides functionalities to create a JSONL file from images in the directory you specify. Images will be captioned using the MiaoshouAI/Florence-2-base-PromptGen-v2.0 model. Images are resized to multiple sizes for better captioning results. [1024, 768, 672, 512].

Why

I wanted to create a tool that would allow me to process images in bulk quickly and efficiently for using in generative art projects. This tool allows me to generate captions for images that I can use as training data captions for my training LORAs (Large OpenAI Research Agents) and other generative models.

Installation (Recommended)

You have a couple of options for installing/running the tool:

Install pipx, then run the tool with the following command

pipx install ecko-cli

Alternatively, you can install using pip

pip install .

Configuration

[!IMPORTANT] Before using the tool, It's required to set up a .env file in the parent directory of the script or your home user dir [windows] or $HOME/.config/ecko-cli-itsjustregi/.env

The application intelligently locates your .env file, accommodating various platforms like Windows and Linux, or defaulting to the current directory.

Usage // Available Commands

Once installed via pipx or pip:

ecko process-images /path/to/images watercolors --padding 4

ecko process-images --use_joy_cap /path/to/images watercolors --padding 4

ecko process-images /path/to/images doors --is_object True

ecko process-images /path/to/images doors --trigger WORD

ecko create-jsonl /path/to/images [dataset]

ecko ui /path/to/images

Dependencies

This tool requires Python 3.11 or higher and has the following dependencies:

"typer",
"rich",
"shellingham",
"python-dotenv",
"torch",
"pillow",
"einops"
"transformers"
"timm"
"torchvision"
"huggingface_hub[cli]"

Contact

For any inquiries, feedback, or suggestions, please feel free to open an issue on this repository.

License

This project is licensed under the MIT License.

---