Add shaded BF16 vs INT4 renders after the first table
#1
by Ramkasuru - opened
- .gitattributes +7 -0
- README.md +11 -0
- assets/sh_renders/020_cube3d_bf16.gif +3 -0
- assets/sh_renders/020_cube3d_int4.gif +3 -0
- assets/sh_renders/118_cube3d_bf16.gif +3 -0
- assets/sh_renders/118_cube3d_int4.gif +0 -0
- assets/sh_renders/191_cube3d_bf16.gif +3 -0
- assets/sh_renders/191_cube3d_int4.gif +3 -0
- assets/sh_renders/305_cube3d_bf16.gif +3 -0
- assets/sh_renders/305_cube3d_int4.gif +3 -0
.gitattributes
CHANGED
|
@@ -36,3 +36,10 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
|
| 36 |
assets/easy.gif filter=lfs diff=lfs merge=lfs -text
|
| 37 |
assets/medium.gif filter=lfs diff=lfs merge=lfs -text
|
| 38 |
assets/complex.gif filter=lfs diff=lfs merge=lfs -text
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 36 |
assets/easy.gif filter=lfs diff=lfs merge=lfs -text
|
| 37 |
assets/medium.gif filter=lfs diff=lfs merge=lfs -text
|
| 38 |
assets/complex.gif filter=lfs diff=lfs merge=lfs -text
|
| 39 |
+
assets/sh_renders/191_cube3d_bf16.gif filter=lfs diff=lfs merge=lfs -text
|
| 40 |
+
assets/sh_renders/191_cube3d_int4.gif filter=lfs diff=lfs merge=lfs -text
|
| 41 |
+
assets/sh_renders/305_cube3d_bf16.gif filter=lfs diff=lfs merge=lfs -text
|
| 42 |
+
assets/sh_renders/305_cube3d_int4.gif filter=lfs diff=lfs merge=lfs -text
|
| 43 |
+
assets/sh_renders/020_cube3d_bf16.gif filter=lfs diff=lfs merge=lfs -text
|
| 44 |
+
assets/sh_renders/020_cube3d_int4.gif filter=lfs diff=lfs merge=lfs -text
|
| 45 |
+
assets/sh_renders/118_cube3d_bf16.gif filter=lfs diff=lfs merge=lfs -text
|
README.md
CHANGED
|
@@ -38,6 +38,17 @@ enabling 3D shape generation on much smaller, more accessible GPUs.
|
|
| 38 |
<mark>💡 The 82% size reduction and 55% VRAM reduction means this model now fits on a single 15 GB GPU (e.g. NVIDIA L4, A10, A2 etc.), bringing high-quality text-to-3D generation to individual researchers and end-user hardware.
|
| 39 |
</mark>
|
| 40 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 41 |
### Original BF16 vs Quantized INT4 Comparisons:
|
| 42 |
##### A. Easy Categories (3)
|
| 43 |

|
|
|
|
| 38 |
<mark>💡 The 82% size reduction and 55% VRAM reduction means this model now fits on a single 15 GB GPU (e.g. NVIDIA L4, A10, A2 etc.), bringing high-quality text-to-3D generation to individual researchers and end-user hardware.
|
| 39 |
</mark>
|
| 40 |
|
| 41 |
+
### Shaded renders: BF16 vs INT4
|
| 42 |
+
|
| 43 |
+
| Prompt | BF16 | **INT4** |
|
| 44 |
+
|---|:-:|:-:|
|
| 45 |
+
| A pineapple with a crown of 8 spiky leaves and a hexagonal surface pattern |  |  |
|
| 46 |
+
| steampunk boombox with 2 circular speakers and 6 brass dials |  |  |
|
| 47 |
+
| A sitting cat with pointed ears and a long tail |  |  |
|
| 48 |
+
| A lace doily pattern extruded to a thin 3D surface |  |  |
|
| 49 |
+
|
| 50 |
+
Rendered with spherical harmonic lighting, Blinn-Phong specular and ambient occlusion: gray only, no color, so the geometry itself shows. The lace doily is a failure case: INT4 loses the lace pattern.
|
| 51 |
+
|
| 52 |
### Original BF16 vs Quantized INT4 Comparisons:
|
| 53 |
##### A. Easy Categories (3)
|
| 54 |

|
assets/sh_renders/020_cube3d_bf16.gif
ADDED
|
Git LFS Details
|
assets/sh_renders/020_cube3d_int4.gif
ADDED
|
Git LFS Details
|
assets/sh_renders/118_cube3d_bf16.gif
ADDED
|
Git LFS Details
|
assets/sh_renders/118_cube3d_int4.gif
ADDED
|
assets/sh_renders/191_cube3d_bf16.gif
ADDED
|
Git LFS Details
|
assets/sh_renders/191_cube3d_int4.gif
ADDED
|
Git LFS Details
|
assets/sh_renders/305_cube3d_bf16.gif
ADDED
|
Git LFS Details
|
assets/sh_renders/305_cube3d_int4.gif
ADDED
|
Git LFS Details
|