search

shenghanlin / SeismicFoundationModel

Seismic Foundation Model
57 stars 12 forks source link

Preparing data for testing #7

Closed HajarAlharthi closed 4 months ago

HajarAlharthi commented 4 months ago

Thank you for sharing your impressive project!

I have a query about testing new data. I've been working with a seismic dataset that includes labels, and I've converted it into .dat files. My intention is to evaluate this data using the fine-tuned model for facies classification, utilizing the visualization notebook. However, the model's outputs didn't match the example output, and I suspect it's because of the data difference. Could you advise me on how did you label/prepare your seismic data or how to properly prepare it for testing with the fine-tuned model? thank you!

shenghanlin commented 4 months ago

Here's a suggested reply to the GitHub query about the mismatch in model outputs due to potential data differences:

Thank you for reaching out and for your interest in the project!

Based on your description, it seems the discrepancy in the model outputs may indeed stem from differences in how the data is prepared or labeled. To address this, I recommend following these steps to fine-tune the model with your specific dataset:

Data Preparation: Ensure that your seismic data is labeled consistently with the model's training dataset. This includes checking the format and the scale of the labels. If there are any discrepancies, you may need to adjust your .dat files to match the format used in the training set. Update DataLoader: Modify your DataLoader to correctly handle your dataset’s specific format and characteristics. This might involve adjusting how you read the .dat files, preprocess the data, and batch the samples. Fine-Tuning: Follow the fine-tuning instructions in the project documentation to fine-tune the model with your data. This step is crucial as it allows the model to adapt to the nuances of your dataset. Model Testing: After fine-tuning, load your model and test it using your validation dataset. This will help you evaluate the model's performance on your data and identify any further adjustments that might be necessary. More details can be found in our paper.

HajarAlharthi commented 3 months ago

Thank you so much for the comprehensive answer, my question would be when using “Facies Identification Challenge: 3-D image interpretation" data. How did you convert it to 2D and did you apply any pre-processing on the data? and why did you use only the In-line or cross line?