mlpc-lab/TokenCompose_SD14_B

🧩 TokenCompose SD14 Model Card

🎬CVPR 2024

TokenCompose_SD14_B is a latent text-to-image diffusion model finetuned from the Stable-Diffusion-v1-4 checkpoint at resolution 512x512 on the VSR split of COCO image-caption pairs for 24,000 steps with a learning rate of 5e-6. The training objective involves token-level grounding terms in addition to denoising loss for enhanced multi-category instance composition and photorealism. The "_A/B" postfix indicates different finetuning runs of the model using the same above configurations.

📄 Paper

Please follow this link.

🧨Example Usage

We strongly recommend using the 🤗Diffuser library to run our model.

import torch
from diffusers import StableDiffusionPipeline

model_id = "mlpc-lab/TokenCompose_SD14_B"
device = "cuda"

pipe = StableDiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.float32)
pipe = pipe.to(device)

prompt = "A cat and a wine glass"
image = pipe(prompt).images[0]  
    
image.save("cat_and_wine_glass.png")

⬆️Improvements over SD14

Method	Multi-category Instance Composition									Photorealism		Efficiency
	Object Accuracy	COCO				ADE20K				FID (COCO)	FID (Flickr30K)	Latency
		MG2	MG3	MG4	MG5	MG2	MG3	MG4	MG5
SD 1.4	29.86	90.721.33	50.740.89	11.680.45	0.880.21	89.810.40	53.961.14	16.521.13	1.890.34	20.88	71.46	7.540.17
TokenCompose (Ours)	52.15	98.080.40	76.161.04	28.810.95	3.280.48	97.750.34	76.931.09	33.921.47	6.210.62	20.19	71.13	7.560.14

📰 Citation

@InProceedings{Wang2024TokenCompose,
    author    = {Wang, Zirui and Sha, Zhizhou and Ding, Zheng and Wang, Yilin and Tu, Zhuowen},
    title     = {TokenCompose: Text-to-Image Diffusion with Token-level Supervision},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2024},
    pages     = {8553-8564}
}