wade3han/champagne

An official codebase for paper ":champagne: CHAMPAGNE: Learning Real-world Conversation from Large-Scale Web Videos (ICCV 23)"

champagne's Issues

[GPU Support] Demo Test

Awesome work and thanks for releasing code! I am working on running champagne on my own nvidia GPU. I wonder if there is any script to run on my own device since it seems that the current code only support to run code on the google cloud. I mean, I just want to inference the model but no train it. Moreover, I notice:

python tpu_run.py --tpu-pod-name ${TPU_NAME} --tpu-size 256 --run-sh ${SCRIPT_SH}

But ${SCRIPT_SH} is not provided, please correct me if there is any misunderstanding.

Best,
Kuma

Missing Files about [unified-io]

Hi Dear Authors!

Thanks you for open-source code!

Since you have shared your pre-trained model, I want to inference it. Could you share any suggestion or script about it? BTW, I know that this paper propose a brand new dataset to further improve the ability of unified-io. So could you tell me how to inference your model?

I notice that '--run_sh' is required for input argument which is not provided in this repo.

Please correct me if I misundertand.

"dialogue" missing

Thank you for releasing your data and code! I downloaded https://storage.googleapis.com/ai2-mosaic-public/projects/champagne/data/ytd_converter_safe_json.tar.gz , but I do not see "dialogue" as a key in the examples in the json. (specifically 00000of3280.json). The website (https://seungjuhan.me/champagne/) seems to indicate that "dialogue" should be one of the keys, and otherwise, I don't see where the transcript is broken up into turns. Is there something I'm missing in how to use the data?