Add load_and_transform_thermal_data

[Abhishekmallik]

Adding the load_and_transform_thermal_data function to data.py

[facebook-github-bot]

Hi @Abhishekmallik!

Thank you for your pull request and welcome to our community.

Action Required

In order to merge any pull request (code, docs, etc.), we require contributors to sign our Contributor License Agreement, and we don't seem to have one on file for you.

Process

In order for us to review and merge your suggested changes, please sign at https://code.facebook.com/cla. If you are contributing on behalf of someone else (eg your employer), the individual CLA may not be sufficient and your employer may need to sign the corporate CLA.

Once the CLA is signed, our tooling will perform checks and validations. Afterwards, the pull request will be tagged with CLA signed. The tagging process may take up to 1 hour after signing. Please give it that time before contacting us about it.

If you have received this in error or have any questions, please contact us at [email protected]. Thanks!

[zhang-ziang]

Thanks for the code!
I find something maybe wrong.
As the thermal data is not the normal image, I think the transforms.Normalize() in transforms.Compose[] should be removed.
I'm recurrencing the metrics on LLVIP dataset, and I find that when the Normalize is added, the binary classification accuracy is about 55.2 and when removed, acc is about 62.4. The latter is more consistent with the paper of ImageBind.

[Jade999]

Thanks for the code! I find something maybe wrong. As the thermal data is not the normal image, I think the transforms.Normalize() in transforms.Compose[] should be removed. I'm recurrencing the metrics on LLVIP dataset, and I find that when the Normalize is added, the binary classification accuracy is about 55.2 and when removed, acc is about 62.4. The latter is more consistent with the paper of ImageBind.

@ZHANG-ZIAN hello, you mean we need to delete transforms.Normalize operation in transforms.Compose，may I ask if this has been confirmed?

[zhang-ziang]

Thanks for the code! I find something maybe wrong. As the thermal data is not the normal image, I think the transforms.Normalize() in transforms.Compose[] should be removed. I'm recurrencing the metrics on LLVIP dataset, and I find that when the Normalize is added, the binary classification accuracy is about 55.2 and when removed, acc is about 62.4. The latter is more consistent with the paper of ImageBind.

@ZHANG-ZIAN hello, you mean we need to delete transforms.Normalize operation in transforms.Compose，may I ask if this has been confirmed?

Maybe. I'm not sure because when I use this method to preprocess the LLVIP test set, the thermal-image retrieval performance is not good as I expect, as ImageBind is train on thermal-image pairs. I also retrained the thermal trunk follow ImageBind paper, but still can't get very good results.
I also get bad result on depth modality. You can find some discussions in the issue where everyone didn't get good results.
Perhaps it is best to wait for the authors themselves to release these preprocessing function code.

[Jade999]

Okay, thank you very much for your experience. Let's wait for the official code
@facebook-github-bot @Abhishekmallik Please take a look at this issue.

[Parv-Maheshwari]

Hi @Abhishekmallik @rohitgirdhar can you please provide an update on the dataloader for thermal.