SS.md

ShapeStacks Dataset

In this document, we provide a step-by-step instruction on:

• How to evaluate our pretrained model on the ShapeStacks Dataset.
• How to train those models yourself.

The ShapeStacks dataset is introduced in CVP. We parse the raw data using this code.

1. Evaluation

1.1 Download Our Dataset

You can download our prepared ShapeStacks dataset using the following script:

gdown --id 1FYtwY03U_xg5lU8j1NHZdQXFDMH8Fjsy -O data/ss.zip
unzip data/ss.zip -d data/

The data structure should look like:

data/ss/train  # The ShapeStacks Training set
data/ss/test   # The ShapeStacks Testing set, containing videos wiht 3 stacked blocks
data/ss/ss4    # The ShapeStacks Testing set, containing videos with 4 stacked blocks

---

1.2 Evaluate Our Prediction Model

You can download our pre-trained RPIN model using the following script:

gdown --id 1VufPAnn2uSeAe1I9KA-NctpvGTjuLscX -O outputs/phys/ss/rpcin.zip
unzip outputs/phys/ss/rpcin.zip -d outputs/phys/ss/

Run the following for evaluation:

sh scripts/test_pred.sh ss rpcin_vae rpcin ${GPU_ID}

2. Training

You can train your prediction model by:

python train.py --cfg configs/ss/rpcin.yaml --gpus ${GPU_ID} --output ${OUTPUT_NAME}