Amanpreet Singh

@yasminabelhadj You can try out Google's Cloud Vision API for OCR text extraction.

Hi @CCYChongyanChen, Yes, we are aware of this. Would you be willing to add newer version of VizWiz to MMF? For VinVL features, we don't have any immediate plans but...

Hi, this is probably happening because you are running out of memory. FastText requires at least 8GB of memory to be loaded. Also, try setting `training.num_workers=0` to spawn less data...

Thanks @TownWilliam for the suggestions. Those will definitely help. You can also try batch size of 16, which should be able to give reasonable results for M4C on TextVQA. On...

This needs to be done at the processor level. Create a process similar to fasttext processor and then replace the config of textvqa dataset to use that processor.

Hi, The FLAVA codebase is on track to be released via torchmultimodal library. I will reply back to this issue by end of this week with further instructions.

Try decreasing your num_workers. It might be too many for a single GPU which means most of the time is spent loading the data.

Confirmed on a separate thread that author is running this on colab. So, that is why it is not picking it up. You will not use `from mmf_cli.run import run`...

Apologies for late reply. MMF actually has a subset split feature. For future reference, please follow instructions in https://github.com/facebookresearch/mmf/commit/637a5bcd201864bdbb42b65fcbc55785e6ba22d6

@steph-en-m Yes, go ahead.