Update README.md
Browse files
README.md
CHANGED
@@ -140,6 +140,10 @@ Each material in the dataset comes with rich metadata, including information on
|
|
140 |
and, when available, descriptions and physical size.
|
141 |
This comprehensive metadata facilitates precise material selection and usage, catering to the specific needs of users.
|
142 |
|
|
|
|
|
|
|
|
|
143 |
## π Dataset Structure
|
144 |
|
145 |
The MatSynth dataset is divided into two splits: the test split, containing 89 materials, and the train split, consisting of 3,980 materials.
|
@@ -182,14 +186,32 @@ presented as the length of the edge in centimeters (358).
|
|
182 |
## π§βπ» Usage
|
183 |
|
184 |
MatSynth is accessible through the datasets python library.
|
|
|
185 |
|
186 |
```python
|
187 |
from datasets import load_dataset
|
|
|
188 |
|
|
|
189 |
ds = load_dataset(
|
190 |
"gvecchio/MatSynth",
|
191 |
streaming = True,
|
192 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
193 |
```
|
194 |
|
195 |
## π Citation
|
|
|
140 |
and, when available, descriptions and physical size.
|
141 |
This comprehensive metadata facilitates precise material selection and usage, catering to the specific needs of users.
|
142 |
|
143 |
+
<center>
|
144 |
+
<img src="https://gvecchio.com/matsynth/static/images/data.png" style="border-radius:10px">
|
145 |
+
</center>
|
146 |
+
|
147 |
## π Dataset Structure
|
148 |
|
149 |
The MatSynth dataset is divided into two splits: the test split, containing 89 materials, and the train split, consisting of 3,980 materials.
|
|
|
186 |
## π§βπ» Usage
|
187 |
|
188 |
MatSynth is accessible through the datasets python library.
|
189 |
+
Following a usage example:
|
190 |
|
191 |
```python
|
192 |
from datasets import load_dataset
|
193 |
+
from torch.utils.data import DataLoader
|
194 |
|
195 |
+
# load the dataset in streaming mode
|
196 |
ds = load_dataset(
|
197 |
"gvecchio/MatSynth",
|
198 |
streaming = True,
|
199 |
)
|
200 |
+
|
201 |
+
# remove unnecessary columns to reduce downloaded data
|
202 |
+
ds = ds.remove_columns(["diffuse", "specular", "displacement", "opacity", "blend_mask"])
|
203 |
+
# keep only specified columns
|
204 |
+
ds = ds.select_columns(["basecolor", "normal", "roughness", " metallic"])
|
205 |
+
|
206 |
+
# filter data matching a specific criteria, e.g.: only CC0 materials
|
207 |
+
ds = ds.filter(lambda x: x["metadata"]["license"] == "CC0")
|
208 |
+
|
209 |
+
# shuffle data
|
210 |
+
ds = ds.shuffle(buffer_size=100)
|
211 |
+
|
212 |
+
# set format for usage in torch
|
213 |
+
ds = ds.with_format("torch")
|
214 |
+
dl = DataLoader(ds, batch_size=8)
|
215 |
```
|
216 |
|
217 |
## π Citation
|