Pose-Aware Embedding Networks and Multi-Modal Image-Language Retrieval

The provided code is intended to support the exploration of embedding spaces, and multi-modal embedding spaces, especially for language. A summary may be found here, and the full document may be found here.

The dataset and pretrained models are available for download.

This code is written in python 3.7.0, and uses PyTorch 0.4.1.

Setting up

Install the python dependencies pip install -r requirements.txt.
Download the dataset, and unzip.
Unzip images_single.zip.
Set the path_dataset variable in the path_manager.py.example file.
Raname path_manager.py.example to path_manager.py.

Training from Scratch

To train and evalute these pose-aware embeddings:

Run the desired model experiments/*/train.py script.
Once complete, run experiments/*/embed.py.
Once complete, run the desired evaluation script in experiments/*/eval/.

To train and evaluate the language mode:

Run the experiments/language/train.py.
Once complete, run experiments/language/embed.py.
Once complete, run the desired evaluation script in experiments/language/eval/.

To produce and evaluate the conditional posebytes:

Run the experiments/language_conditional/generate_single_question_posebyte.py.
Once complete, run experiments/language_conditional/embed.py.
Once complete, run the desired evaluation script in experiments/language_conditional/eval/.

To train and evaluate the conditional masks:

Run the experiments/language_masks/train.py.
Once complete, run the desired evaluation script in experiments/language_masks/eval/.

Evaluation Only

Evaluating the pose-aware embeddings:

Copy the desired precomputed_embeddings/image/*/embedding_0.py to the corresponding experiments/image/*/embeddings folder.
Run the desired evaulation located in the respective eval folder.

Evaluating the language mode:

Copy the precomputed_embeddings/language/embedding_valtest_0.py to the corresponding experiments/language embeddings folder.
Run the desired evaulation located in the respective eval folder.

Evaluating the conditional posebytes:

Copy the precomputed_embeddings/language_conditional/embedding_conditional.py to the corresponding experiments/language_conditional/ embeddings folder.
Run the desired evaulation located in the respective eval folder.

Evaluating the pose-aware masks:

Copy the contents of precomputed_masks/ to the corresponding experiments/language_mask/masks/ folder.
Run the desired evaulation located in the respective eval folder.

DCurro/Pose-Aware-Embedding-Networks-and-Multi-Modal-Image-Language-Retrieval

About

Resources

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Languages

Packages