Diversity-Algorithm-CLI-Tool

CLI tool to Diversity Algorithm (Noob Friendly)

Simply run the da-picker.py to index your data with the following arguments:

• --INPUT_FILE_PATH : Path to PyTorch Model
• --DATA_PATH : Path to Dataset
• --UMAP : Enable UMAP for Diversity Embeddings
• --subset_size : Percentage of points in the entire dataset to selected by the Diversity Algorithm
• --OUTPUT_FILE_PATH : File path for the Diversity Algorithm Output. Creates a new dataset using Diverse Embeddings.
• --metric : Prints out a Class Distribution
• --img_size : Size of Images (Required when passing model)
• --embedding_size: Size of model's output embedding (Required when passing model)

Update

removed CSV, FAISS, PKL processing utils. Feel free to get them from the other branch.

Example:

!python /content/Diversity-Algorithm-CLI-Tool/da-picker.py --DATA_PATH /content/UCMerced_LandUse/Images --img_size 64 --embedding_size 2048 --UMAP True --subset_size 0.1  --INPUT_FILE_PATH "/content/pt_model.pt" --OUTPUT_FILE_PATH "/content/embedding_path_output" --technique DA_FAST --sample_size 0.5 --metric True

Dependencies

Install the necessary packages from requirements.txt using pip install -r requirements.txt before you run the scripts.

TODO

• Add Pt Model Support
• Add DA Variant
• Code Restructuring
• Tf Model Support
• Package
• Multiprocessing