Poster

Install Requirement

conda create -n project python=3.10
conda activate project
pip install "cython<3.0.0" && pip install --no-build-isolation pyyaml==5.4.1
pip install -r requirements.txt

Checkpoint Preparation

• In case there is no checkpoint zipped together with this code, please download the checkpoint from this link.

• Unzip the file and the checkpoint is ready for use.

Data Preparation for LDM

Run the following commands in order:

python load_data.py

python data_processing.py

python data/prepare_latent.py\
    --gpu 0 \
    --config configs/vae/config_nodisc_kl1e-6_64.yaml \
    --ckpt ckpt/train_vae_11-22-212308_cond_64_nos2c_w1e-1/epoch=238-step=504768.ckpt

Training

VAE Training

CUDA_VISIBLE_DEVICES=0 \
python train_vae.py \
    --config configs/vae/config_nodisc_kl1e-6_64.yaml \
    --exp_name cond_64_nos2c_w1e-1

LDM Training

CUDA_VISIBLE_DEVICES=0 \
python train_ldm.py \
    --config configs/diffusion/ldm_cond_64_cfg0.0.yaml \
    --exp_name cond_64_cfg0.0 

Inference

VAE Inference (Not important) Example command to get reconstruciton data for val and test sets:

CUDA_VISIBLE_DEVICES=0 \
python inference_vae.py \
    --config configs/vae/config_nodisc_kl1e-6_64.yaml \
    --exp_name cond_64_airplane \
    --ckpt ckpt/train_vae_11-22-212308_cond_64_nos2c_w1e-1/epoch=238-step=504768.ckpt \
    --target_categories <category>

<category> is selected in [ "table", "chair", "airplane" ]

To get metric measurement for reconstruction

python eval.py airplane output/vae_reconstruction/cond_64_airplane/rec_data.npy

LDM Inference - IMPORTANT Command line to sample 1000 voxels for each category

CUDA_VISIBLE_DEVICES=0 \
python inference_ldm.py \
    --config ./configs/diffusion/ldm_cond_64_cfg0.0.yaml \
    --ckpt ckpt/ldm_11-27-205036_cond_64_JSD_cfg0.0/epoch=399-step=26400.ckpt \
    --output_dir ldm/cond_64_JSD_cfg0.0_ep399 \
    --target_category <category>

<category> is selected in [ "table", "chair", "airplane" ]

Command lines to get quantitative measurement:

python eval.py chair output/ldm/cond_64_JSD_cfg0.0_ep399/chair/samples.npy

python eval.py table output/ldm/cond_64_JSD_cfg0.0_ep399/table/samples.npy

python eval.py airplane output/ldm/cond_64_JSD_cfg0.0_ep399/airplane/samples.npy