search

caoyunkang / Segment-Any-Anomaly

Official implementation of "Segment Any Anomaly without Training via Hybrid Prompt Regularization (SAA+)".
704 stars 76 forks source link

custom dataset #23

Closed sulaiman1988 closed 10 months ago

sulaiman1988 commented 1 year ago

How to implement a repo on custom datasets to get segmentation for the image by image? How to Finetune using a custom dataset?

PauloFavero commented 1 year ago

I also have the same questions

caoyunkang commented 1 year ago

Hello, thank you for your interest in SAA+.

To integrate this repository with your specific dataset on an individual image basis, I recommend consulting the ./demo.py file. This script provides guidance on processing individual images. It's important to configure the textual_prompts and property_text_prompts based on the characteristics of your dataset. This enables SAA+ to accurately identify anomalies within your images. If your intention is to apply this to a custom dataset, you can start by examining the datasets directory. Begin by creating a custom.py file. Subsequently, refer to SAA/prompts to document your tailored prompts.

With regards to fine-tuning, I'd like to clarify that our SAA+ is designed to operate without requiring any additional training to accommodate new data. As of now, fine-tuning is not supported by our system. If you have any further inquiries, please don't hesitate to ask.