Spaces:
Runtime error
Runtime error
yuanze1024
commited on
Commit
•
04ea559
1
Parent(s):
55e7aed
update app
Browse files- Dockerfile +1 -1
- app.py +31 -21
Dockerfile
CHANGED
@@ -22,7 +22,7 @@ RUN pip install --no-cache-dir --upgrade -r /code/requirements.txt
|
|
22 |
# note that you may need to modify the TORCH_CUDA_ARCH_LIST in the setup.py file
|
23 |
ENV TORCH_CUDA_ARCH_LIST="8.6"
|
24 |
|
25 |
-
# Install Pointnet2_PyTorch
|
26 |
RUN git clone https://github.com/yuanze1024/Pointnet2_PyTorch.git && cd Pointnet2_PyTorch/pointnet2_ops_lib && pip install .
|
27 |
|
28 |
COPY --chown=user:user . /code
|
|
|
22 |
# note that you may need to modify the TORCH_CUDA_ARCH_LIST in the setup.py file
|
23 |
ENV TORCH_CUDA_ARCH_LIST="8.6"
|
24 |
|
25 |
+
# Install Pointnet2_PyTorch, pip install git+ won't work for unknown reason
|
26 |
RUN git clone https://github.com/yuanze1024/Pointnet2_PyTorch.git && cd Pointnet2_PyTorch/pointnet2_ops_lib && pip install .
|
27 |
|
28 |
COPY --chown=user:user . /code
|
app.py
CHANGED
@@ -89,13 +89,12 @@ def retrieve_3D_models(textual_query, top_k, modality_list):
|
|
89 |
indices = _retrieve_3D_models(textual_query, top_k, modality_list)
|
90 |
return [get_image_and_id(index) for index in indices]
|
91 |
|
92 |
-
def get_sub_dataset(sub_dataset_id):
|
93 |
"""
|
94 |
get sub-dataset by sub_dataset_id [1, 1000]
|
95 |
|
96 |
Returns:
|
97 |
caption: str
|
98 |
-
difficulty: str
|
99 |
images: list of tuple (PIL.Image, str)
|
100 |
"""
|
101 |
rel = relation[sub_dataset_id - 1]
|
@@ -111,18 +110,23 @@ def get_sub_dataset(sub_dataset_id):
|
|
111 |
return new_image
|
112 |
|
113 |
results = []
|
114 |
-
|
115 |
-
|
116 |
-
|
117 |
-
|
118 |
-
|
119 |
-
|
|
|
|
|
|
|
|
|
|
|
120 |
|
121 |
-
return caption,
|
122 |
|
123 |
-
def feel_lucky():
|
124 |
sub_dataset_id = random.randint(1, 1000)
|
125 |
-
return sub_dataset_id, *get_sub_dataset(sub_dataset_id)
|
126 |
|
127 |
def launch():
|
128 |
with gr.Blocks() as demo: # https://sketchfab.com/3d-models/fd30f87848c9454c9225eccc39726787
|
@@ -131,14 +135,17 @@ def launch():
|
|
131 |
with gr.Tab("Retrieval Visualization"):
|
132 |
with gr.Row():
|
133 |
md2 = gr.Markdown(r"""### Visualization for Text-Based-3D Model Retrieval
|
134 |
-
We build a visualization demo to demonstrate the text-based-3D model retrievals. Due to the memory limitation of HF Space,
|
|
|
|
|
135 |
|
136 |
**Note**:
|
137 |
|
138 |
-
The *Modality List* refers to the features ensembled by the retrieval methods. According to our experiment results, basically the more modalities, the better performance the methods gets.
|
|
|
|
|
139 |
with gr.Row():
|
140 |
-
textual_query = gr.Textbox(label="Textual Query", autofocus=True,
|
141 |
-
placeholder="A chair with a wooden frame and a cushioned seat")
|
142 |
modality_list = gr.CheckboxGroup(label="Modality List", value=[],
|
143 |
choices=["text", "front", "back", "left", "right", "above",
|
144 |
"below", "diag_above", "diag_below", "3D"])
|
@@ -173,19 +180,22 @@ Here is a visualization of the dataset.
|
|
173 |
|
174 |
**Note:**
|
175 |
|
176 |
-
The *Query* is used in this sub-dataset. The *
|
177 |
-
The color surrounding the 3D model indicates whether it is a good fit for the textual query.
|
|
|
178 |
with gr.Row():
|
179 |
lucky = gr.Button("I'm Feeling Lucky !", scale=1, variant='primary')
|
180 |
-
query_id = gr.Number(label="Sub-dataset ID", scale=1, minimum=1, maximum=1000, step=1, interactive=True)
|
|
|
181 |
query = gr.Textbox(label="Textual Query", scale=3, interactive=False)
|
182 |
-
difficulty = gr.Textbox(label="Query Difficulty", scale=1, interactive=False)
|
183 |
# model3d = gr.Model3D(interactive=False, scale=1)
|
184 |
with gr.Row():
|
185 |
output2 = gr.Gallery(format="webp", label="3D Models in Sub-dataset", columns=5, type="pil", interactive=False)
|
186 |
|
187 |
-
lucky.click(feel_lucky, outputs=[query_id, query,
|
188 |
-
query_id.submit(get_sub_dataset, query_id, [query,
|
|
|
189 |
|
190 |
demo.queue(max_size=10)
|
191 |
demo.launch(server_name='0.0.0.0')
|
|
|
89 |
indices = _retrieve_3D_models(textual_query, top_k, modality_list)
|
90 |
return [get_image_and_id(index) for index in indices]
|
91 |
|
92 |
+
def get_sub_dataset(sub_dataset_id, sorted=False):
|
93 |
"""
|
94 |
get sub-dataset by sub_dataset_id [1, 1000]
|
95 |
|
96 |
Returns:
|
97 |
caption: str
|
|
|
98 |
images: list of tuple (PIL.Image, str)
|
99 |
"""
|
100 |
rel = relation[sub_dataset_id - 1]
|
|
|
110 |
return new_image
|
111 |
|
112 |
results = []
|
113 |
+
if not sorted:
|
114 |
+
for ind in target_ids:
|
115 |
+
image, source_id = get_image_and_id(source_to_id[ind])
|
116 |
+
results.append((handle_image(image, True if ind in GT_ids else False), source_id))
|
117 |
+
else:
|
118 |
+
for gt_id in GT_ids:
|
119 |
+
image, source_id = get_image_and_id(source_to_id[gt_id])
|
120 |
+
results.append((handle_image(image, True), source_id))
|
121 |
+
for neg_id in negative_ids:
|
122 |
+
image, source_id = get_image_and_id(source_to_id[neg_id])
|
123 |
+
results.append((handle_image(image, False), source_id))
|
124 |
|
125 |
+
return caption, results
|
126 |
|
127 |
+
def feel_lucky(is_sorted):
|
128 |
sub_dataset_id = random.randint(1, 1000)
|
129 |
+
return sub_dataset_id, *get_sub_dataset(sub_dataset_id, is_sorted)
|
130 |
|
131 |
def launch():
|
132 |
with gr.Blocks() as demo: # https://sketchfab.com/3d-models/fd30f87848c9454c9225eccc39726787
|
|
|
135 |
with gr.Tab("Retrieval Visualization"):
|
136 |
with gr.Row():
|
137 |
md2 = gr.Markdown(r"""### Visualization for Text-Based-3D Model Retrieval
|
138 |
+
We build a visualization demo to demonstrate the text-based-3D model retrievals. Due to the memory limitation of HF Space,
|
139 |
+
we only support the [Uni3D](https://github.com/baaivision/Uni3D) which has shown an excellent performance in our benchmark.
|
140 |
+
What's more, **we only search in a subset of Objaverse, which contains 89K 3D models**.
|
141 |
|
142 |
**Note**:
|
143 |
|
144 |
+
The *Modality List* refers to the features ensembled by the retrieval methods. According to our experiment results, basically the more modalities, the better performance the methods gets.
|
145 |
+
|
146 |
+
Also, you may want to ckeck the 3D model in a 3D model viewer, in that case, you can visit [Objaverse](https://objaverse.allenai.org/explore) for exploration.""")
|
147 |
with gr.Row():
|
148 |
+
textual_query = gr.Textbox(label="Textual Query", autofocus=True, value="Super Mario")
|
|
|
149 |
modality_list = gr.CheckboxGroup(label="Modality List", value=[],
|
150 |
choices=["text", "front", "back", "left", "right", "above",
|
151 |
"below", "diag_above", "diag_below", "3D"])
|
|
|
180 |
|
181 |
**Note:**
|
182 |
|
183 |
+
The *Query* is used in this sub-dataset. The *Sorted* will put the Ground Truths in the front of the results.
|
184 |
+
The color surrounding the 3D model indicates whether it is a good fit for the textual query.
|
185 |
+
A **<span style="color:#00FF00">green</span>** color suggests a Ground Truth, while a **<span style="color:#FF0000">red</span>** color indicates a mismatch.""")
|
186 |
with gr.Row():
|
187 |
lucky = gr.Button("I'm Feeling Lucky !", scale=1, variant='primary')
|
188 |
+
query_id = gr.Number(label="Sub-dataset ID", scale=1, minimum=1, maximum=1000, step=1, interactive=True, value=986)
|
189 |
+
is_sorted = gr.Checkbox(value=False, label="", scale=1, info="Sorted")
|
190 |
query = gr.Textbox(label="Textual Query", scale=3, interactive=False)
|
191 |
+
# difficulty = gr.Textbox(label="Query Difficulty", scale=1, interactive=False)
|
192 |
# model3d = gr.Model3D(interactive=False, scale=1)
|
193 |
with gr.Row():
|
194 |
output2 = gr.Gallery(format="webp", label="3D Models in Sub-dataset", columns=5, type="pil", interactive=False)
|
195 |
|
196 |
+
lucky.click(feel_lucky, inputs=is_sorted, outputs=[query_id, query, output2])
|
197 |
+
query_id.submit(get_sub_dataset, [query_id, is_sorted], [query, output2])
|
198 |
+
is_sorted.change(get_sub_dataset, [query_id, is_sorted], [query, output2])
|
199 |
|
200 |
demo.queue(max_size=10)
|
201 |
demo.launch(server_name='0.0.0.0')
|