shap-e icon indicating copy to clipboard operation
shap-e copied to clipboard

How to improve the quality of 3d model?

Open realcarlos opened this issue 2 years ago • 4 comments

I tried txt_2_3d and image_2_3d, the model is rough,

doraemon

Screenshot from 2023-05-16 19-43-41

mario

Screenshot from 2023-05-16 19-42-13

house

Screenshot from 2023-05-16 19-43-01

man

Screenshot from 2023-05-16 19-41-51

realcarlos avatar May 16 '23 11:05 realcarlos

I think it is possible to generate 3D models from multi-view images to improve accuracy, rather than just one image. I used the following code to try to load three views in the project deployed by Google gloab, but the result failed.

`batch_size = 1 guidance_scale = 3.0

要读取的多个图片文件路径

image_paths = ["test1.png", "test2.png","test3.png"]

读取图片并存储在一个列表中

images = [load_image(path) for path in image_paths]

latents = sample_latents( batch_size=batch_size, model=model, diffusion=diffusion, guidance_scale=guidance_scale, model_kwargs=dict(images=images * batch_size), # 将图片列表传递给 images 参数 progress=True, clip_denoised=True, use_fp16=True, use_karras=True, karras_steps=64, sigma_min=1e-3, sigma_max=160, s_churn=0, ) `

Saktawdi avatar May 16 '23 14:05 Saktawdi

Use it with Zero 123 model its used for generating multiple views you can try it there: https://huggingface.co/spaces/cvlab/zero123-live

xaranex1 avatar May 16 '23 14:05 xaranex1

I think it is possible to generate 3D models from multi-view images to improve accuracy, rather than just one image. I used the following code to try to load three views in the project deployed by Google gloab, but the result failed.

`batch_size = 1 guidance_scale = 3.0

要读取的多个图片文件路径

image_paths = ["test1.png", "test2.png","test3.png"]

读取图片并存储在一个列表中

images = [load_image(path) for path in image_paths]

latents = sample_latents( batch_size=batch_size, model=model, diffusion=diffusion, guidance_scale=guidance_scale, model_kwargs=dict(images=images * batch_size), # 将图片列表传递给 images 参数 progress=True, clip_denoised=True, use_fp16=True, use_karras=True, karras_steps=64, sigma_min=1e-3, sigma_max=160, s_churn=0, ) `

It looks that the image300M model does not support multiple-views.

yuedajiong avatar Jun 09 '23 03:06 yuedajiong

I think it is possible to generate 3D models from multi-view images to improve accuracy, rather than just one image. I used the following code to try to load three views in the project deployed by Google gloab, but the result failed. `batch_size = 1 guidance_scale = 3.0

要读取的多个图片文件路径

image_paths = ["test1.png", "test2.png","test3.png"]

读取图片并存储在一个列表中

images = [load_image(path) for path in image_paths] latents = sample_latents( batch_size=batch_size, model=model, diffusion=diffusion, guidance_scale=guidance_scale, model_kwargs=dict(images=images * batch_size), # 将图片列表传递给 images 参数 progress=True, clip_denoised=True, use_fp16=True, use_karras=True, karras_steps=64, sigma_min=1e-3, sigma_max=160, s_churn=0, ) `

It looks that the image300M model does not support multiple-views.

i was facing same issue, so which model supports multiple images?

devanghingu avatar Jul 07 '23 16:07 devanghingu