GDR Grouped Discrete Representation for Object-Centric Learning

⚗️ (2026/01/06) Update !!!

Please check our brand new OCL works:

• RandSF.Q: significantly surpasses state-of-the-art video OCL, e.g., SlotContrast, by up to 10 points!
• SmoothSA: improves the state of the art even further, e.g., SPOT / DIAS (images) and SlotContrast / RandSF.Q (videos), with minimal modifications!

---

About

Official implementation of ECML 2025 paper "Grouped Discrete Representation for Object-Centric Learning" available on arXiv:2411.02299.

Please note that GDR is re-implemented upon codebase 🤗 VQ-VFM-OCL, different from the version described in the paper. For more details, models, checkpoints, datasets and results, please visit this repo.

Quantitative results of object discovery on COCO: (Encoding with backbone DINO2/S-14 at resolution 256x256/224x224)

Converted Datasets 🚀

Dataset COCO is available on dataset-coco, which is converted in LMDB database format and can be used off-the-shelf in this repo.

Model Checkpoints 🌟

The checkpoints for the models are available.

• slate-gdr-coco: SLATE-GDR, i.e., GDR-Tfd, on COCO.
• slotdiffusion-gdr-coco: SlotDiffusion-GDR, i.e., GDR-Dfz, on COCO.

How to Use

Take SLATE-GDR on COCO as an example.

(1) Environment

To set up the environment, run:

# python 3.11
pip install -r requirements.txt

(2) Dataset

To prepare the dataset, download Converted Datasets and unzip to path/to/your/dataset/. Or convert them by yourself according to XxxDataset.convert_dataset() docs.

(3) Train

To train the model, run:

# 1. pretrain the GDR VAE module
python train.py \
    --seed 42 \
    --cfg_file config-slatesteve/vqvae-coco-c256-gdr.py \
    --data_dir path/to/your/dataset \
    --save_dir save

# *. place the best VAE checkpoint at archive-slatesteve/vqvae-coco-c256-gdr/best.pth
mv save archive-slatesteve

# 2. train the SLATE-GDR OCL model
python train.py \
    --seed 42 \
    --cfg_file config-slatesteve/slate_r_vqvae-coco-gdr.py \
    --data_dir path/to/your/dataset \
    --save_dir save \
    --ckpt_file archive-slatesteve/vqvae-coco-c256-gdr/best.pth

(4) Evaluate

To evaluate the model, run:

python eval.py \
    --cfg_file config-slatesteve/slate_r_vqvae-coco-gdr.py \
    --data_dir path/to/your/dataset \
    --ckpt_file archive-slatesteve/slate_r_vqvae-coco-gdr/best.pth \
    --is_viz True
# object discovery accuracy values will be printed in the terminal
# object discovery visualization will be saved to ./slate_r_vqvae-coco-gdr/

Support

If you have any issues on this repo or cool ideas on OCL, please do not hesitate to contact me!

• page: https://genera1z.github.io
• email: [email protected], [email protected]

Citation

If you find this repo useful, please cite our work.

@article{zhao2025gdr,
  title={{Grouped Discrete Representation for Object-Centric Learning}},
  author={Zhao, Rongzhen and Wang, Vivienne and Kannala, Juho and Pajarinen, Joni},
  journal={ECML PKDD},
  year={2025}
}