From 4e887cafb15eabf8559d599978238ba294a0c1d8 Mon Sep 17 00:00:00 2001 From: Daniel Tedesco Date: Wed, 23 Nov 2022 14:11:53 +0800 Subject: [PATCH] Update grammar/spelling and some references. --- README.md | 26 ++++++++++++++------------ 1 file changed, 14 insertions(+), 12 deletions(-) diff --git a/README.md b/README.md index 0b6a8424..bd426eff 100644 --- a/README.md +++ b/README.md @@ -1,20 +1,22 @@ # Dreambooth on Stable Diffusion -This is an implementtaion of Google's [Dreambooth](https://arxiv.org/abs/2208.12242) with [Stable Diffusion](https://github.com/CompVis/stable-diffusion). The original Dreambooth is based on [Imagen](https://imagen.research.google/) text-to-image model. However, neither the model nor the pre-trained weights of Imagen is available. To enable people to fine-tune a text-to-image model with a few examples, I implemented the idea of Dreambooth on Stable diffusion. +This is an implementation of Google's [Dreambooth](https://arxiv.org/abs/2208.12242) with [Stable Diffusion](https://github.com/CompVis/stable-diffusion). The original Dreambooth is based on the [Imagen](https://imagen.research.google/) text-to-image model. However, neither the model nor the pre-trained weights of Imagen is publicly available. To enable people to fine-tune a text-to-image model with a few examples, I implemented the idea of Dreambooth on Stable diffusion. -This code repository is based on that of [Textual Inversion](https://github.com/rinongal/textual_inversion). Note that Textual Inversion only optimizes word ebedding, while dreambooth fine-tunes the whole diffusion model. +This code repository is based on that of [Textual Inversion](https://github.com/rinongal/textual_inversion). Note that Textual Inversion only optimizes word embedding, while dreambooth fine-tunes the whole diffusion model. + +The implementation makes minimum changes over the official codebase of Textual Inversion. In fact, due to laziness, some components in Textual Inversion, such as the embedding manager, are not deleted, although they will never be used here. -The implementation makes minimum changes over the official codebase of Textual Inversion. In fact, due to lazyness, some components in Textual Inversion, such as the embedding manager, are not deleted, although they will never be used here. ## Update -**9/20/2022**: I just found a way to reduce the GPU memory a bit. Remember that this code is based on Textual Inversion, and TI's code base has [this line](https://github.com/rinongal/textual_inversion/blob/main/ldm/modules/diffusionmodules/util.py#L112), which disable gradient checkpointing in a hard-code way. This is because in TI, the Unet is not optimized. However, in Dreambooth we optimize the Unet, so we can turn on the gradient checkpoint pointing trick, as in the original SD repo [here](https://github.com/CompVis/stable-diffusion/blob/main/ldm/modules/diffusionmodules/util.py#L112). The gradient checkpoint is default to be True in [config](https://github.com/XavierXiao/Dreambooth-Stable-Diffusion/blob/main/configs/stable-diffusion/v1-finetune_unfrozen.yaml#L47). I have updated the codes. +**9/20/2022**: I just found a way to reduce the GPU memory a bit. Remember that this code is based on Textual Inversion, and TI's code base has [this line](https://github.com/rinongal/textual_inversion/blob/main/ldm/modules/diffusionmodules/util.py#L112), which disables gradient checkpointing in a hard-code way. This is because in TI, the Unet is not optimized. However, in Dreambooth we optimize the Unet, so we can enable the gradient checkpoint pointing trick, as in the original SD repo [here](https://github.com/CompVis/stable-diffusion/blob/main/ldm/modules/diffusionmodules/util.py#L112). The gradient checkpoint defaults to True in [config](https://github.com/XavierXiao/Dreambooth-Stable-Diffusion/blob/main/configs/stable-diffusion/v1-finetune_unfrozen.yaml#L47). + ## Usage ### Preparation -First set-up the ```ldm``` enviroment following the instruction from textual inversion repo, or the original Stable Diffusion repo. +First set up the ```ldm``` enviroment, following instructions from textual inversion repo or the original Stable Diffusion repo. -To fine-tune a stable diffusion model, you need to obtain the pre-trained stable diffusion models following their [instructions](https://github.com/CompVis/stable-diffusion#stable-diffusion-v1). Weights can be downloaded on [HuggingFace](https://huggingface.co/CompVis). You can decide which version of checkpoint to use, but I use ```sd-v1-4-full-ema.ckpt```. +To fine-tune a stable diffusion model, you need to obtain the pre-trained stable diffusion models ([SDv1 instructions](https://github.com/CompVis/stable-diffusion#stable-diffusion-v1)). Weights can be downloaded on [HuggingFace](https://huggingface.co/CompVis). You can decide which version of checkpoint to use, but I use ```sd-v1-4-full-ema.ckpt```. -We also need to create a set of images for regularization, as the fine-tuning algorithm of Dreambooth requires that. Details of the algorithm can be found in the paper. Note that in the original paper, the regularization images seem to be generated on-the-fly. However, here I generated a set of regularization images before the training. The text prompt for generating regularization images can be ```photo of a ```, where `````` is a word that describes the class of your object, such as ```dog```. The command is +We also need to create a set of images for regularization, a requirement for fine-tuning Dreambooth's algorithm. Details of the algorithm can be found in [the paper](https://arxiv.org/abs/2208.12242). Note that in the original paper, the regularization images seem to be generated on-the-fly. However, here I generated a set of regularization images before training. The text prompt for generating regularization images can be ```photo of a ```, where `````` is a word that describes the class of your object, such as ```dog```. The command follows the format: ``` python scripts/stable_txt2img.py --ddim_eta 0.0 --n_samples 8 --n_iter 1 --scale 10.0 --ddim_steps 50 --ckpt /path/to/original/stable-diffusion/sd-v1-4-full-ema.ckpt --prompt "a photo of a " @@ -28,7 +30,7 @@ We should definitely use more images for regularization. Please try 100 or 200, For some cases, if the generated regularization images are highly unrealistic (happens when you want to generate "man" or "woman"), you can find a diverse set of images (of man/woman) online, and use them as regularization images. ### Training -Training can be done by running the following command +Training can be done by running the command: ``` python main.py --base configs/stable-diffusion/v1-finetune_unfrozen.yaml @@ -38,12 +40,12 @@ python main.py --base configs/stable-diffusion/v1-finetune_unfrozen.yaml --gpus 0, --data_root /root/to/training/images --reg_data_root /root/to/regularization/images - --class_word + --class_word ``` -Detailed configuration can be found in ```configs/stable-diffusion/v1-finetune_unfrozen.yaml```. In particular, the default learning rate is ```1.0e-6``` as I found the ```1.0e-5``` in the Dreambooth paper leads to poor editability. The parameter ```reg_weight``` corresponds to the weight of regularization in the Dreambooth paper, and the default is set to ```1.0```. +Detailed configuration can be found in ```configs/stable-diffusion/v1-finetune_unfrozen.yaml```. In particular, the default learning rate is ```1.0e-6```, as I found the ```1.0e-5``` in the Dreambooth paper leads to poor editability. The parameter ```reg_weight``` corresponds to the weight of regularization in the Dreambooth paper, and the default is set to ```1.0```. -Dreambooth requires a placeholder word ```[V]```, called identifier, as in the paper. This identifier needs to be a relatively rare tokens in the vocabulary. The original paper approaches this by using a rare word in T5-XXL tokenizer. For simplicity, here I just use a random word ```sks``` and hard coded it.. If you want to change that, simply make a change in [this file](https://github.com/XavierXiao/Dreambooth-Stable-Diffusion/blob/main/ldm/data/personalized.py#L10). +Dreambooth requires a placeholder word ```[V]```, called an identifier in the paper. This identifier needs to be a relatively rare token in the vocabulary. The original paper approaches this by using a rare word in its T5-XXL tokenizer. For simplicity, here I just use a random word ```sks``` and hard coded it. If you want to change that, simply update [this file](https://github.com/XavierXiao/Dreambooth-Stable-Diffusion/blob/main/ldm/data/personalized.py#L10). Training will be run for 800 steps, and two checkpoints will be saved at ```./logs//checkpoints```, one at 500 steps and one at final step. Typically the one at 500 steps works well enough. I train the model use two A6000 GPUs and it takes ~15 mins. @@ -63,7 +65,7 @@ python scripts/stable_txt2img.py --ddim_eta 0.0 In particular, ```sks``` is the identifier, which should be replaced by your choice if you happen to change the identifier, and `````` is the class word ```--class_word``` for training. ## Results -Here I show some qualitative results. The training images are obtained from the [issue](https://github.com/rinongal/textual_inversion/issues/8) in the Textual Inversion repository, and they are 3 images of a large trash container. Regularization images are generated by prompt ```photo of a container```. Regularization images are shown here: +Here I show some qualitative results. The training images are obtained from the [issue](https://github.com/rinongal/textual_inversion/issues/8) in the Textual Inversion repository, and they are 3 images of a large trash container. Regularization images are generated by the prompt ```photo of a container```. Regularization images are shown here: ![](assets/a-container-0038.jpg)