ahaliassos

Results 8 comments of ahaliassos

Add Defocus and ZoomBlur transforms

Hi, any news on when these two transforms will be added to albumentations?

Cannot reproduce AUC on FF++ test split.

Hi, The provided checkpoint was trained on FF++ with c23 (HQ) compression, so the HQ results should match. I'm not sure why they do not currently. Do the final mouth...

How are multiple faces handled in preprocessing ?

Hi, Indeed, the code assumes that each video contains one face that needs to be extracted. For example, in FF++, only the largest face is tracked and extracted (see Appendix...

Path building error in stage2/eval.py

Does `{working directory}/outputs/YYYY-MM-DD/HH-MM-SS/stage2/eval.py` exist. And are you sure you're not running eval.py from that directory rather than `{working directory}/stage2/eval.py`?

The configuration file for the large model has incorrect parameters, which is causing the model to fail to load correctly.

Hi, thanks for spotting this! You are right, the parameters were wrong. In general, the width of the decoder should match that of the encoder, except in the low-resource setting,...

The configuration file for the large model has incorrect parameters, which is causing the model to fail to load correctly.

Fixed, thank you.

The configuration file for the large model has incorrect parameters, which is causing the model to fail to load correctly.

The high-resource base LRS3+Vox2 weights file was corrupted, but I have now uploaded it again. I also fixed the config for the visual backbone (which was accidentally not pushed to...

Request for tokenizer

Hi, you can have a look at https://github.com/ahaliassos/usr/blob/main/utils/labels/unigram1000_units.txt for the vocabulary and at https://github.com/ahaliassos/usr/blob/main/utils/utils.py#L6 for how the final token list is obtained (i.e., with the blank and eos tokens). Hope...