Update README.md

README.md

@@ -9,16 +9,16 @@ viewer: false
   <img src="contents/vae.png" alt="small" width="800">
 </p>
 
-| original, reconstructed from float16, reconstructed from uint8*
+*original, reconstructed from float16, reconstructed from uint8*
 
 
 Find 138 GB of imagenet dataset too bulky? Did you know entire imagenet actually just fits inside the ram of apple watch?
 
-* Center-croped, resized to 256x256
+* Resized, Center-croped to 256x256
 * VAE compressed with [SDXL's VAE](https://huggingface.co/stabilityai/sdxl-vae)
 * Further quantized to int8 near-lossless manner, compressing the entire training dataset of 1,281,167 images down to just 5GB!
 
-Introducing Imagenet.int8, the new MNIST of 2024. After the great popularity of the Latent Diffusion era (Thank you stable diffusion!), its *almost* the standard to use VAE version of the imagenet for diffusion-model training. As you might know, lot of great diffusion research is based on latent variation of the imagenet. 
+Introducing Imagenet.int8, the new MNIST of 2024. After the great popularity of the [Latent Diffusion](https://arxiv.org/abs/2112.10752) (Thank you stable diffusion!), its *almost* the standard to use VAE version of the imagenet for diffusion-model training. As you might know, lot of great diffusion research is based on latent variation of the imagenet. 
 
 These include: 
 
@@ -29,9 +29,9 @@ These include:
 * [Min-SNR](https://openaccess.thecvf.com/content/ICCV2023/html/Hang_Efficient_Diffusion_Training_via_Min-SNR_Weighting_Strategy_ICCV_2023_paper.html)
 * [MDT](https://openaccess.thecvf.com/content/ICCV2023/papers/Gao_Masked_Diffusion_Transformer_is_a_Strong_Image_Synthesizer_ICCV_2023_paper.pdf)
 
-... but so little material online on the actual preprocessed dataset. I'm here to fix that. One thing I noticed was that latent doesn't have to be full precision! Indeed, they can be as small as int-8, and it won't hurt! Here are some of the examples:
+... but so little material online on the actual preprocessed dataset. I'm here to fix that. One thing I noticed was that latent doesn't have to be full precision! Indeed, they can be as small as int-8, and it doesn't hurt!
 
-So clearly, it doesn't make sense to download entire Imagenet and do VAE everytime. Just download this, `to('cuda')` the entire dataset just to flex, and call it a day.😌
+So clearly, it doesn't make sense to download entire Imagenet and process with VAE everytime. Just download this, `to('cuda')` the entire dataset just to flex, and call it a day.😌
 
 (BTW If you think you'll need higher precision, you can always further fine-tune your model on higher precision. But I doubt that.)
 
@@ -91,12 +91,13 @@ train_dataloader = torch.utils.data.DataLoader(
     batch_size=32,
     num_workers=3,
 )
+```
 
+Thats the dataloader! Now, below is the example usage. Notice how you have to reshape as data is flattened.
 
-
+```
 ###### Example Usage. Let's see if we can get the 5th image. BTW shuffle plz
 
-
 model = "stabilityai/your-stable-diffusion-model"
 vae = AutoencoderKL.from_pretrained("stabilityai/sdxl-vae").to("cuda:0")
 
@@ -115,4 +116,20 @@ print(f"idx: {idx}, text_label: {text_label}, latent: {vae_latent.shape}")
 x = vae.decode(vae_latent.cuda()).sample
 img = VaeImageProcessor().postprocess(image = x.detach(), do_denormalize = [True, True])[0]
 img.save("5th_image.png")
-```
+```
+
+Enjoy!
+
+# Citation
+
+```bibtex
+@misc{imagenet_int8,
+  author       = {Simo Ryu},
+  title        = {Imagenet.int8: Entire Imagenet dataset in 5GB},
+  year         = 2024,
+  publisher    = {Hugging Face Datasets},
+  url          = {https://huggingface.co/datasets/cloneofsimo/imagenet.int8},
+  note         = {Entire Imagenet dataset compressed to 5GB using VAE and quantized with int8}
+}
+```
+