Thank you very much for providing the code implementation!
I have made some adjustments to the DeepInversionFeatureHook and DeepInversionClass in deepinversion.py, to enable multi-gpu support. In this version, the r_feature loss on each device is summed together to compute the total loss -- mean also works but I found that it required some adjustments to the lr.
Thank you very much for providing the code implementation!
I have made some adjustments to the
DeepInversionFeatureHookandDeepInversionClassindeepinversion.py, to enable multi-gpu support. In this version, ther_featureloss on each device is summed together to compute the total loss -- mean also works but I found that it required some adjustments to the lr.