In this directory, we illustrate the details of our experiments on the task of visually grounded story generation.
- 1. Data Preparation
- 2. Language Model Training
- 3. Perform Inference with Magic
- 4. Perform Inference with Baseline Methods
To prepare the data and image index for the task, please follow instructions [here].
To train the language model on the ROCStories benchmark, please follow the instructions [here].
To perform inference with our magic approach, please refer to details [here] and [here].
To perform inference with the contrastive search baseline, please refer to details [here].
To perform inference with the greedy search baseline, please refer to details [here].
To perform inference with the beam search baseline, please refer to details [here].
To perform inference with the top-k sampling baseline, please refer to details [here].
To perform inference with the nucleus sampling baseline, please refer to details [here].
To perform inference with the typical sampling baseline, please refer to details [here].