ControlAR

Controllable Image Generation with Autoregressive Models

Zongming Li^1,*, Tianheng Cheng^1,*, Shoufa Chen², Peize Sun², Haocheng Shen³,Longjin Ran³, Xiaoxin Chen³, Wenyu Liu¹, Xinggang Wang^1,📧

¹ Huazhong University of Science and Technology, ² The University of Hong Kong ³ vivo AI Lab

(* equal contribution, 📧 corresponding author)

News

[2024-12-12]: We introduce control strength factor, employ a larger control encoder(dinov2-base), and optimize text alignment capabilities along with generation diversity. New model weight: depth_base.safetensors and edge_base.safetensors. The edge_base.safetensors can handle three types of edges, including Canny, HED, and Lineart.
[2024-10-31]: The code and models have been released!
[2024-10-04]: We have released the technical report of ControlAR. Code, models, and demos are coming soon!

Highlights

ControlAR explores an effective yet simple conditional decoding strategy for adding spatial controls to autoregressive models, e.g., LlamaGen, from a sequence perspective.
ControlAR supports arbitrary-resolution image generation with autoregressive models without hand-crafted special tokens or resolution-aware prompts.

TODO

release code & models.
release demo code and HuggingFace demo: HuggingFace Spaces 🤗

Results

We provide both quantitative and qualitative comparisons with diffusion-based methods in the technical report!

Models

We released checkpoints of text-to-image ControlAR on different controls and settings, i.e. arbitrary-resolution generation.

AR Model	Type	Control encoder	Control	Arbitrary-Resolution	Checkpoint
LlamaGen-XL	t2i	DINOv2-small	Canny Edge	✅	ckpt
LlamaGen-XL	t2i	DINOv2-small	Depth	✅	ckpt
LlamaGen-XL	t2i	DINOv2-small	HED Edge	❌	ckpt
LlamaGen-XL	t2i	DINOv2-small	Seg. Mask	❌	ckpt
LlamaGen-XL	t2i	DINOv2-base	Edge (Canny, Hed, Lineart)	❌	ckpt
LlamaGen-XL	t2i	DINOv2-base	Depth	❌	ckpt

Getting Started

Installation

conda create -n ControlAR python=3.10
git clone https://github.com/hustvl/ControlAR.git
cd ControlAR
pip install torch==2.1.2+cu118 --extra-index-url https://download.pytorch.org/whl/cu118
pip install -r requirements.txt
pip3 install -U openmim 
mim install mmengine 
mim install "mmcv==2.1.0"
pip3 install "mmsegmentation>=1.0.0"
pip3 install mmdet
git clone https://github.com/open-mmlab/mmsegmentation.git

Pretrained Checkpoints for ControlAR

tokenizer	text encoder	LlamaGen-B	LlamaGen-L	LlamaGen-XL
vq_ds16_t2i.pt	flan-t5-xl	c2i_B_256.pt	c2i_L_256.pt	t2i_XL_512.pt

We recommend storing them in the following structures:

|---checkpoints
      |---t2i
            |---canny/canny_MR.safetensors
            |---hed/hed.safetensors
            |---depth/depth_MR.safetensors
            |---seg/seg_cocostuff.safetensors
            |---edge_base.safetensors
            |---depth_base.safetensors
      |---t5-ckpt
            |---flan-t5-xl
                  |---config.json
                  |---pytorch_model-00001-of-00002.bin
                  |---pytorch_model-00002-of-00002.bin
                  |---pytorch_model.bin.index.json
                  |---tokenizer.json
      |---vq
            |---vq_ds16_c2i.pt
            |---vq_ds16_t2i.pt
      |---llamagen (Only necessary for training)
            |---c2i_B_256.pt
            |---c2i_L_256.pt
            |---t2i_XL_stage2_512.pt

Demo

Coming soon...

Sample & Generation

1. Class-to-image genetation

python autoregressive/sample/sample_c2i.py \
--vq-ckpt checkpoints/vq/vq_ds16_c2i.pt \
--gpt-ckpt checkpoints/c2i/canny/LlamaGen-L.pt \
--gpt-model GPT-L --seed 0 --condition-type canny

2. Text-to-image generation

Generate an image using HED edge and text-to-image ControlAR:

python autoregressive/sample/sample_t2i.py \
--vq-ckpt checkpoints/vq/vq_ds16_t2i.pt \
--gpt-ckpt checkpoints/t2i/hed/hed.safetensors \
--gpt-model GPT-XL --image-size 512 \
--condition-type hed --seed 0 --condition-path condition/example/t2i/multigen/eye.png

Generate an image using segmentation mask and text-to-image ControlAR:

python autoregressive/sample/sample_t2i.py \
--vq-ckpt checkpoints/vq/vq_ds16_t2i.pt \
--gpt-ckpt checkpoints/t2i/seg/seg_cocostuff.safetensors \
--gpt-model GPT-XL --image-size 512 \
--condition-type seg --seed 0 --condition-path condition/example/t2i/cocostuff/doll.png \
--prompt 'A stuffed animal wearing a mask and a leash, sitting on a pink blanket'

3. Text-to-image generation with adjustable control strength

Generate an image using depth map and text-to-image ControlAR:

python autoregressive/sample/sample_t2i.py \
--vq-ckpt checkpoints/vq/vq_ds16_t2i.pt \
--gpt-ckpt checkpoints/t2i/depth_base.safetensors \
--gpt-model GPT-XL --image-size 512 \
--condition-type seg --seed 0 --condition-path condition/example/t2i/multigen/bird.jpg \
--prompt 'A bird made of blue crystal' \
--adapter-size base \
--control-strength 0.6

Generate an image using lineart edge and text-to-image ControlAR:

python autoregressive/sample/sample_t2i.py \
--vq-ckpt checkpoints/vq/vq_ds16_t2i.pt \
--gpt-ckpt checkpoints/t2i/edge_base.safetensors \
--gpt-model GPT-XL --image-size 512 \
--condition-type lineart --seed 0 --condition-path condition/example/t2i/multigen/girl.jpg \
--prompt 'A girl with blue hair' \
--adapter-size base \
--control-strength 0.6

4. Arbitrary-resolution generation

python3 autoregressive/sample/sample_t2i_MR.py --vq-ckpt checkpoints/vq/vq_ds16_t2i.pt \
--gpt-ckpt checkpoints/t2i/depth_MR.safetensors --gpt-model GPT-XL --image-size 768 \
--condition-type depth --condition-path condition/example/t2i/multi_resolution/bird.jpg \
--prompt 'colorful bird' --seed 0

python3 autoregressive/sample/sample_t2i_MR.py --vq-ckpt checkpoints/vq/vq_ds16_t2i.pt \
--gpt-ckpt checkpoints/t2i/canny_MR.safetensors --gpt-model GPT-XL --image-size 768 \
--condition-type canny --condition-path condition/example/t2i/multi_resolution/bird.jpg \
--prompt 'colorful bird' --seed 0

Preparing Datasets

We provide the dataset datails for evaluation and training. If you don't want to train ControlAR, just download the validation splits.

1. Class-to-image

Download ImageNet and save it to data/imagenet/data.

2. Text-to-image

Download ADE20K with caption(~7GB) and save the .parquet files to data/Captioned_ADE20K/data.
Download COCOStuff with caption( ~62GB) and save the .parquet files to data/Captioned_COCOStuff/data.
Download MultiGen-20M( ~1.22TB) and save the .parquet files to data/MultiGen20M/data.

3. Preprocessing datasets

To save training time, we adopt the tokenizer to pre-process the images with the text prompts.

ImageNet

bash scripts/autoregressive/extract_file_imagenet.sh \
--vq-ckpt checkpoints/vq/vq_ds16_c2i.pt \
--data-path data/imagenet/data/val \
--code-path data/imagenet/val/imagenet_code_c2i_flip_ten_crop \
--ten-crop --crop-range 1.1 --image-size 256

ADE20k

bash scripts/autoregressive/extract_file_ade.sh \
--vq-ckpt checkpoints/vq/vq_ds16_t2i.pt \
--data-path data/Captioned_ADE20K/data --code-path data/Captioned_ADE20K/val \
--ten-crop --crop-range 1.1 --image-size 512 --split validation

COCOStuff

bash scripts/autoregressive/extract_file_cocostuff.sh \
--vq-ckpt checkpoints/vq/vq_ds16_t2i.pt \
--data-path data/Captioned_COCOStuff/data --code-path data/Captioned_COCOStuff/val \
--ten-crop --crop-range 1.1 --image-size 512 --split validation

MultiGen

bash scripts/autoregressive/extract_file_multigen.sh \
--vq-ckpt checkpoints/vq/vq_ds16_t2i.pt \
--data-path data/MultiGen20M/data --code-path data/MultiGen20M/val \
--ten-crop --crop-range 1.1 --image-size 512 --split validation

Testing and Evaluation

1. Class-to-image generation on ImageNet

bash scripts/autoregressive/test_c2i.sh \
--vq-ckpt ./checkpoints/vq/vq_ds16_c2i.pt \
--gpt-ckpt ./checkpoints/c2i/canny/LlamaGen-L.pt \
--code-path /path/imagenet/val/imagenet_code_c2i_flip_ten_crop \
--gpt-model GPT-L --condition-type canny --get-condition-img True \
--sample-dir ./sample --save-image True

python create_npz.py --generated-images ./sample/imagenet/canny

Then download imagenet validation data which contains 10000 images, or you can use the whole validation data as reference data by running val.sh.

Calculate the FID score:

python evaluations/c2i/evaluator.py /path/imagenet/val/FID/VIRTUAL_imagenet256_labeled.npz \
sample/imagenet/canny.npz

2. Text-to-image generation on ADE20k

Download Mask2Former(weight) and save it to evaluations/.

Use this command to get 2000 images based on the segmentation mask:

bash scripts/autoregressive/test_t2i.sh --vq-ckpt checkpoints/vq/vq_ds16_t2i.pt \
--gpt-ckpt checkpoints/t2i/seg/seg_ade20k.pt \
--code-path data/Captioned_ADE20K/val --gpt-model GPT-XL --image-size 512 \
--sample-dir sample/ade20k --condition-type seg --seed 0

Calculate mIoU of the segmentation masks from the generated images:

python evaluations/ade20k_mIoU.py

3. Text-to-image generation on COCOStuff

Download DeepLabV3(weight) and save it to evaluations/.

Generate images using segmentation masks as condition controls:

bash scripts/autoregressive/test_t2i.sh --vq-ckpt checkpoints/vq/vq_ds16_t2i.pt \
--gpt-ckpt checkpoints/t2i/seg/seg_cocostuff.pt \
--code-path data/Captioned_COCOStuff/val --gpt-model GPT-XL --image-size 512 \
--sample-dir sample/cocostuff --condition-type seg --seed 0

Calculate mIoU of the segmentation masks from the generated images:

python evaluations/cocostuff_mIoU.py

4. Text-to-image generation on MultiGen-20M

We adopt generation with HED edges as the example:

Generate 5000 images based on the HED edges generated from validation images

bash scripts/autoregressive/test_t2i.sh --vq-ckpt checkpoints/vq/vq_ds16_t2i.pt \
--gpt-ckpt checkpoints/t2i/hed/hed.safetensors --code-path data/MultiGen20M/val \
--gpt-model GPT-XL --image-size 512 --sample-dir sample/multigen/hed \
--condition-type hed --seed 0

Evaluate the conditional consistency (SSIM):

python evaluations/hed_ssim.py

Calculate the FID score:

python evaluations/clean_fid.py --val-images data/MultiGen20M/val/image --generated-images sample/multigen/hed/visualization

Training ControlAR

1. Class-to-image (Canny)

bash scripts/autoregressive/train_c2i_canny.sh --cloud-save-path output \
--code-path data/imagenet/train/imagenet_code_c2i_flip_ten_crop \
--image-size 256 --gpt-model GPT-B --gpt-ckpt checkpoints/llamagen/c2i_B_256.pt

2. Text-to-image (Canny)

bash scripts/autoregressive/train_t2i_canny.sh

Acknowledgments

The development of ControlAR is based on LlamaGen, ControlNet, ControlNet++, and AiM, and we sincerely thank the contributors for thoese great works!

Citation

If you find ControlAR is useful in your research or applications, please consider giving us a star 🌟 and citing it by the following BibTeX entry.

@article{li2024controlar,
      title={ControlAR: Controllable Image Generation with Autoregressive Models}, 
      author={Zongming Li, Tianheng Cheng, Shoufa Chen, Peize Sun, Haocheng Shen, Longjin Ran, Xiaoxin Chen, Wenyu Liu, Xinggang Wang},
      year={2024},
      eprint={2410.02705},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2410.02705}, 
}

Name		Name	Last commit message	Last commit date
Latest commit History 32 Commits
assets		assets
autoregressive		autoregressive
condition		condition
dataset		dataset
demo		demo
evaluations		evaluations
language		language
scripts		scripts
tokenizer		tokenizer
tools		tools
utils		utils
.DS_Store		.DS_Store
.gitignore		.gitignore
LICENSE		LICENSE
README.md		README.md
create_npz.py		create_npz.py
requirements.txt		requirements.txt

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

ControlAR

Controllable Image Generation with Autoregressive Models

News

Highlights

TODO

Results

Models

Getting Started

Installation

Pretrained Checkpoints for ControlAR

Demo

Sample & Generation

1. Class-to-image genetation

2. Text-to-image generation

3. Text-to-image generation with adjustable control strength

4. Arbitrary-resolution generation

Preparing Datasets

1. Class-to-image

2. Text-to-image

3. Preprocessing datasets

Testing and Evaluation

1. Class-to-image generation on ImageNet

2. Text-to-image generation on ADE20k

3. Text-to-image generation on COCOStuff

4. Text-to-image generation on MultiGen-20M

Training ControlAR

1. Class-to-image (Canny)

2. Text-to-image (Canny)

Acknowledgments

Citation

About

Releases

Packages

Contributors 2

Languages

License

hustvl/ControlAR

Folders and files

Latest commit

History

Repository files navigation

ControlAR

Controllable Image Generation with Autoregressive Models

News

Highlights

TODO

Results

Models

Getting Started

Installation

Pretrained Checkpoints for ControlAR

Demo

Sample & Generation

1. Class-to-image genetation

2. Text-to-image generation

3. Text-to-image generation with adjustable control strength

4. Arbitrary-resolution generation

Preparing Datasets

1. Class-to-image

2. Text-to-image

3. Preprocessing datasets

Testing and Evaluation

1. Class-to-image generation on ImageNet

2. Text-to-image generation on ADE20k

3. Text-to-image generation on COCOStuff

4. Text-to-image generation on MultiGen-20M

Training ControlAR

1. Class-to-image (Canny)

2. Text-to-image (Canny)

Acknowledgments

Citation

About

Resources

License

Stars

Watchers

Forks

Releases

Packages 0

Contributors 2

Languages

Packages