LDong

How to divide the dataset into train/val/test?

1

Hi, I'd like to know how to divide the dataset into train/val/test. It seems to be divided as train/val=8:1. Is this true?

Hi, I'd like to know how to get the audio portion of the HowTo100M Dataset, as the [HowTo100 website](https://www.di.ens.fr/willow/research/howto100m/) showed that the audio portion has been removed.

loading bytes object failed.

1

Hi, I used the command used in [Memory file -> numpy -> memory file #156](https://github.com/kkroening/ffmpeg-python/issues/156#issue-393302441), ``` input_data = client.get() # bytes object decode = ( ffmpeg .input('pipe:') .output('pipe:') .get_args() )...

Why is my count for the number of videos in htm_aa_v1.csv 247, 564 instead of 370K?

4

Hi, I downloaded htm_aa_v1.csv from the [Oxford](http://www.robots.ox.ac.uk/~htd/tan/htm_aa_v1.csv) server you given, I used np.unique to count the video list and found only 247, 564 videos but not 370k videos. Is there...

Hi, could you release the code for evaluation on HTM-AA?

Stupid automatic update of pytorch when pip install torchtext

2

## ❓ Questions and Help I installed torchtext using pip and it automatically downloaded torchtext==1.13.1, but it automatically updated my pytorch==1.12.1 to pytorch=1.13.1 without considering that my torchvision and torchaudio...

Hi, Could you share the codes of linear probing AVID model on ESC50 and DCASE dataset? There are only codes for fine-tuning AVID model on HMDB51 and UCF101 dataset and...

Hi, I installed the package following your guidance. However, when I ran the demo_video.py, it raised a ValueError: Unpack failed: incomplete input when doing **state=checkpoints.restore_checkpoint(ckpt_dir_path, target_state, step=step, prefix='ckpt_', parallel=True)** in...

Could you release the trained checkpoint?

1

Hi, could you release the trained checkpoint for eval only?