README-infer.md

Inference Guide

This README provides instructions for running the inference script after completing stage two of the training process.

Overview

The inference script processes images from a specified dataset folder and generates new images based on the trained model. This is typically done after stage two of the training pipeline has been completed.

Usage

To run the inference process, use the following command:

Replace FOLDER_NAME with the same folder name used during stage two training.

# Specify the folder's name directly in the script
folder_name="0"  # Replace with your desired input folder

# Run the inference script with the specified arguments
python inference.py \
    --dataset_path datasets/$folder_name \
    --model_path outputs/$folder_name \
    --output_path generated_images/ \
    --intrinsic_anchors texture material color

Notes

1. Ensure the folder_name variable matches the folder name used during stage two training.

2. [Important] The --intrinsic_anchors parameter must be identical to what was specified during stage two training (in this example: "texture material color").

3. Generated images will be saved to the generated_images/ directory.

4. If the inference results after running the model aren't satisfactory, try experimenting with different values for --reg_intrinsic_weight (try 1, 5e-2, or 5e-4, etc) or change the --seed value. These adjustments can often lead to better outcomes.

Back to Main Documentation

For complete system documentation and information about all stages of the ICE pipeline, please refer to the main README document.