Update app.py
Browse files
app.py
CHANGED
@@ -57,6 +57,17 @@ The prompt we use for every video is "A robotic arm with a gripper and a small c
|
|
57 |
"""
|
58 |
|
59 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
60 |
def create_key(seed=0):
|
61 |
return jax.random.PRNGKey(seed)
|
62 |
|
@@ -300,8 +311,12 @@ def create_demo(process, max_images=12, default_num_images=4):
|
|
300 |
format="avi",
|
301 |
interactive=False)
|
302 |
|
303 |
-
|
304 |
-
|
|
|
|
|
|
|
|
|
305 |
|
306 |
|
307 |
|
|
|
57 |
"""
|
58 |
|
59 |
|
60 |
+
perfo_description = """
|
61 |
+
The Table on the right shows the performances of our models running on different nodes.
|
62 |
+
To make the benchmark, we loaded one of our model on every GPUs of the node. We then retrieve an episode of our simulation.
|
63 |
+
For every frame of the episode, we preprocess the image (resize, canny, ...) and process the Canny image on the GPUs.
|
64 |
+
We repeated this procedure for different Batch Size (BS).
|
65 |
+
|
66 |
+
We can see that the greater the BS the greater the FPS. By increazing the BS, we make a profit on the parallelization of the GPUs.
|
67 |
+
|
68 |
+
"""
|
69 |
+
|
70 |
+
|
71 |
def create_key(seed=0):
|
72 |
return jax.random.PRNGKey(seed)
|
73 |
|
|
|
311 |
format="avi",
|
312 |
interactive=False)
|
313 |
|
314 |
+
with gr.Row():
|
315 |
+
with gr.Column():
|
316 |
+
gr.Markdown(perfo_description)
|
317 |
+
with gr.Column():
|
318 |
+
gr.Image("./perfo_rtx.png",
|
319 |
+
interactive=False)
|
320 |
|
321 |
|
322 |
|