This repo contains downloading instructions for the two OpenViDial datasets in:

• OpenViDial 1.0: OpenViDial: A Large-Scale, Open-Domain Dialogue Dataset with Visual Contexts
• OpenViDial 2.0: OpenViDial 2.0: A Larger-Scale, Open-Domain Dialogue Generation Dataset with Visual Contexts

and the code to reproduce results based on the two OpenViDial datasets in the paper Modeling Text-visual Mutual Dependency for Multi-modal dialog Generation

When humans converse, what a speaker will say next significantly depends on what he sees. OpenViDial is a largescale multi-module dialogue dataset for this purpose. Thes dialogue turns and visual contexts are extracted from movies and TV series, where each dialogue turn is paired with the corresponding visual context in which it takes place. Up to 2022.01.22 OpenViDial has two verseion: OpenViDial 1.0 and OpenViDial 2.0. For OpenViDial 1.0, it contains a total number of 1.1 million dialogue turns, and thus 1.1 million visual contexts stored in images. For OpenViDial 2.0, it is much larger than the previous version OpenViDial 1.0 containing a total number of 5.6 million dialogue turns along with 5.6 million visual contexts stored in images.

The following are two short conversations where visual contexts are crucial.

For the detailed and downloading instructions for the two OpenViDial datasets (OpenViDial 1.0, OpenViDial 2.0) can be found here

If you'd like to take a glance at the a sample of the dataset instead of downloading the full dataset, we provide a data sample here. The small size data are sampled from OpenViDial 1.0 dataset, and can be used for debug or any other operations.

We proposed three models for this dataset. Please refer to the paper for details:

• Model #1 - NoVisual: use only dialog texts without visual information

• Model #2 - CoarseVisual: use texts and a pretrained ResNet50 on ImageNet to compute 1000-d feature from each picture

• Model #3 - FineVisual: use texts and a pretrained Faster R-CNN on Genome to compute 2048-d * K objects features from each picture

• python >= 3.6
• pip install -r requirements.txt

preprocessed_data_dir is a directory that contains all the preprocessed files (text, image feature mmap, offsets, etc.) generated from origin_data_dir and we use them in training models. The directory structure is shown below.

Note: every train* file or directory should have a 'valid' and a 'test' counterpart, we ignore them below for simplicity.

├──preprocessed_data_dir
      └── train.features.mmap  // numpy mmap array file of shape [num_sents, 1000], each row is a 1000-d ResNet-50 feature
      └── train.objects.mmap  // numpy mmap array file of shape [num_sents, 20, 2048],  faster-rcnn object feature file, each row contain 20 objects feature, which is 2048-d
      └── train.objects_mask.mmap  // numpy mmap array file of shape [num_sents, 20],  faster-rcnn mask file, each row contain 20 objects mask, 1 for valid, 0 for mask
      └── train.offsets.npy  // numpy array file of shape [num_episodes], each item is the offsets of one episode
      └── train.sent_num.npy // numpy array file of shape [num_episodes], each item is the sentence number of one episode

We use Moses Tokenizer to tokenize texts and generate offsets arrays: bash ./scripts/preprocess_video_data.sh and followed with byte-pair-encoding and fairseq-preprocess binarization: bash ./scripts/preprocess_text_data.sh

Note: You need to change DATA_DIR, ORIGIN_DIR, OUTPUT_DIR to your own path.

CNN-pooling features is used for Model #2 - CoarseVisual and Faster R-CNN features is used for Model #3 - FineVisual. You can directly download the pre-computed files for CNN and Faster-RCNN features here for either OpenViDial 1.0 dataset or OpenViDial 2.0 dataset.

If you want to extract some feature on your own, or you'd like to know details of extracting visual features, see video_dialogue_model/extract_features/extract_features.md

Note: Extracting features will take you too much time.

bash scripts/reproduce_baselines/text_only.sh will train and evaluate NoVisual, Remember to change MODEL_DIR and DATA_DIR for your setup.

Note: fairseq may use all gpus on your machine and the actual batch size is times by number of gpus. Therefore, if you use multiple gpus, batch size should be devided by number of gpus.

bash scripts/reproduce_baselines/text_and_img_feature.sh will train and evaluate CoarseVisual. Remember to change MODEL_DIR and DATA_DIR for your setup. Please make sure you use one single gpu to reproduce our results.

bash scripts/reproduce_baselines/text_and_img_objects.sh will train and evaluate FineVisual, Remember to change MODEL_DIR and DATA_DIR for your setup. Please make sure you use one single gpu to reproduce our results.

For NV seeing ./mmi/text/README.md. The structure of training data used in both CV and FV is same as the former part.

bash ./mmi/text/train.sh && bash ./mmi/text/mmi_generate.sh will train and evaluate MI-NV. Remember to change all the MODEL_DIR and DATA_DIR for your setup. Please make sure you use one signle gpu to reproduce our results.

bash ./mmi/feature/scrtpts/train_image.sh && bash ./mmi/feature/scrtpts/mmi_feature_generate.sh will train and evaluate MI-CV. Remember to change all the MODEL_DIR and DATA_DIR for your setup. Please make sure you use one signle gpu to reproduce our results.

bash ./mmi/feature/scrtpts/train_object.sh && bash ./mmi/feature/scrtpts/mmi_object_generate.sh will train and evaluate MI-FV. Remember to change all the MODEL_DIR and DATA_DIR for your setup. Please make sure you use one signle gpu to reproduce our results.

• get diversity statistics of system output: train/stats.py
• get rouge statistics of system output: train/rouge.py

The size of OpenViDial 2.0 dataset is too much larger than that of OpenViDial 1.0 dataset. To make the results reproducibility we didn't use the all features for CoarseVisual and FineVisual model (only 5% in this experiments), since the full features will occupy too much memory and may not avaliable for most researchers.