Update app.py

app.py

@@ -18,8 +18,8 @@ toxicity = Detoxify('original')
 # Constants
 ####################################
 AVAILABLE_MODELS = {
-    'TripoSR': 'TripSR',
-    'Shape-E': 'shap-e',
+    'TripoSR': 'TripoSR',
+    'LGM-Mini': 'LGM-Mini',
 }
 
 SPACE_ID = os.getenv('SPACE_ID')
@@ -171,15 +171,15 @@ css = ".generating {visibility: hidden}"
 
 model_names = {
     'TripoSR': 'TripoSR',
-    'Shap-E': 'Shap-E',
+    'LGM-Mini': 'LGM-Mini',
 }
 model_licenses = {
     'TripoSR': 'MIT License',
-    'Shap-E': 'MIT License'
+    'LGM-Mini': 'MIT License'
 }
 model_links = {
     'TripoSR': 'https://github.com/VAST-AI-Research/TripoSR',
-    'Shap-E': 'https://github.com/openai/shap-e',
+    'LGM-Mini': 'https://github.com/3DTopia/LGM',
 }
 
 def model_license(name):
@@ -312,68 +312,32 @@ with gr.Blocks() as leaderboard:
     leaderboard.load(get_leaderboard, inputs=[reveal_prelim], outputs=[df])
     reloadbtn.click(get_leaderboard, inputs=[reveal_prelim], outputs=[df])
     
-def synthandreturn(text):
-    text = text.strip()
-    if len(text) > MAX_SAMPLE_TXT_LENGTH:
-        raise gr.Error(f'You exceeded the limit of {MAX_SAMPLE_TXT_LENGTH} characters')
-    if len(text) < MIN_SAMPLE_TXT_LENGTH:
-        raise gr.Error(f'Please input a text longer than {MIN_SAMPLE_TXT_LENGTH} characters')
-    if (
-        # test toxicity 
-        toxicity.predict(text)['toxicity'] > 0.8
-    ):
-        print(f'Detected toxic content! "{text}"')
-        raise gr.Error('Your text failed the toxicity test')
-    if not text:
-        raise gr.Error(f'You did not enter any text')
-    # Check language
-    try:
-        if not detect(text) == "en":
-            gr.Warning('Warning: The input text may not be in English')
-    except:
-        pass
+def synthandreturn(image):
+    
     # Get two random models
     mdl1, mdl2 = random.sample(list(AVAILABLE_MODELS.keys()), 2)
-    log_text(text)
     print("[debug] Using", mdl1, mdl2)
-    def predict_and_update_result(text, model, result_storage):
+    def predict_and_update_result(image, model, result_storage):
         try:
             if model in AVAILABLE_MODELS:
-                result = router.predict(text, AVAILABLE_MODELS[model].lower(), api_name="/synthesize")
+                result = router.predict(image, AVAILABLE_MODELS[model].lower(), api_name="/synthesize")
             else:
-                result = router.predict(text, model.lower(), api_name="/synthesize")
+                result = router.predict(image, model.lower(), api_name="/synthesize")
         except:
             raise gr.Error('Unable to call API, please try again :)')
         print('Done with', model)
-        try:
-            with tempfile.NamedTemporaryFile(suffix='.wav', delete=False) as f:
-                audio = AudioSegment.from_file(result)
-                current_sr = audio.frame_rate
-                if current_sr > 24000:
-                    audio = audio.set_frame_rate(24000)
-                try:
-                    print('Trying to normalize audio')
-                    audio = match_target_amplitude(audio, -20)
-                except:
-                    print('[WARN] Unable to normalize audio')
-                audio.export(f.name, format="wav")
-                os.unlink(result)
-                result = f.name
-        except:
-            pass
-
         result_storage[model] = result
     
     results = {}
-    thread1 = threading.Thread(target=predict_and_update_result, args=(text, mdl1, results))
-    thread2 = threading.Thread(target=predict_and_update_result, args=(text, mdl2, results))
+    thread1 = threading.Thread(target=predict_and_update_result, args=(image, mdl1, results))
+    thread2 = threading.Thread(target=predict_and_update_result, args=(image, mdl2, results))
     thread1.start()
     thread2.start()
     thread1.join()
     thread2.join()
     
     return (
-        text,
+        image,
         "Synthesize",
         gr.update(visible=True), # r2
         mdl1, # model1
@@ -395,24 +359,24 @@ with gr.Blocks() as vote:
     gr.Markdown(INSTR)
     with gr.Group():
         with gr.Row():
-            text = gr.Textbox(container=False, show_label=False, placeholder="Enter text to synthesize", lines=1, max_lines=1, scale=9999999, min_width=0)
+            image = gr.Image(container=False, show_label=False,  min_width=0)
         btn = gr.Button("Synthesize", variant='primary')
     model1 = gr.Textbox(interactive=False, lines=1, max_lines=1, visible=False)
     model2 = gr.Textbox(interactive=False, lines=1, max_lines=1, visible=False)
     with gr.Row(visible=False) as r2:
         with gr.Column():
             with gr.Group():
-                aud1 = gr.Audio(interactive=False, show_label=False, show_download_button=False, show_share_button=False, waveform_options={'waveform_progress_color': '#3C82F6'})
+                threed1 = gr.Model3D(show_label=False, show_download_button=True, show_share_button=False)
                 abetter = gr.Button("A is better", variant='primary')
                 prevmodel1 = gr.Textbox(interactive=False, show_label=False, container=False, value="Vote to reveal model A", text_align="center", lines=1, max_lines=1, visible=False)
         with gr.Column():
             with gr.Group():
-                aud2 = gr.Audio(interactive=False, show_label=False, show_download_button=False, show_share_button=False, waveform_options={'waveform_progress_color': '#3C82F6'})
+                threed2 = gr.Model3D(show_label=False, show_download_button=True, show_share_button=False)
                 bbetter = gr.Button("B is better", variant='primary')
                 prevmodel2 = gr.Textbox(interactive=False, show_label=False, container=False, value="Vote to reveal model B", text_align="center", lines=1, max_lines=1, visible=False)
     nxtroundbtn = gr.Button('Next round', visible=False)
-    outputs = [text, btn, r2, model1, model2, aud1, aud2, abetter, bbetter, prevmodel1, prevmodel2, nxtroundbtn]
-    btn.click(synthandreturn, inputs=[text], outputs=outputs)
+    outputs = [image, btn, r2, model1, model2, threed1, threed2, abetter, bbetter, prevmodel1, prevmodel2, nxtroundbtn]
+    btn.click(synthandreturn, inputs=[image], outputs=outputs)
     nxtroundbtn.click(clear_stuff, outputs=outputs)