:sparkles: update description

README.md

@@ -15,7 +15,9 @@ tags:
   <img src="https://huggingface.co/KokeCacao/mvdream-hf/resolve/main/doc/image_3.png" height="256">
 </p>
 
-A huggingface implementation of MVDream, used for quick one-line download. See [huggingface repo](https://huggingface.co/KokeCacao/mvdream-hf/tree/main) that hosts sd-v1.5 version. See [github repo](https://github.com/KokeCacao/mvdream-hf) for convertion code.
+A huggingface implementation of MVDream with 4 views, used for quick one-line download. See [huggingface repo](https://huggingface.co/KokeCacao/mvdream-hf/tree/main) that hosts sd-v1.5 version and [huggingface repo](https://huggingface.co/KokeCacao/mvdream-base-hf) for sd-v2.1 version. See [github repo](https://github.com/KokeCacao/mvdream-hf) for convertion code.
+
+Note that the original paper presents the sd-v2.1 version. Images above are generated with sd-v2.1 version.
 
 ## Convert Original Weights to Diffusers
 

scripts/convert_mvdream_to_diffusers.py

@@ -6,14 +6,8 @@ import sys
 
 sys.path.insert(0, '../')
 
-from transformers import (
-    CLIPImageProcessor,
-    CLIPVisionModelWithProjection,
-)
-
 from diffusers.models import (
     AutoencoderKL,
-    UNet2DConditionModel,
 )
 from diffusers.schedulers import DDIMScheduler
 from diffusers.utils import logging