Update app.py

app.py

@@ -3,6 +3,7 @@ import os
 import random
 import numpy as np
 import gdown
+import base64
 from time import gmtime, strftime
 from csv import writer
 
@@ -39,13 +40,26 @@ else:  # local data
     VIDEO_PATH = 'robotinder-data'
 
 
+def inference(video_path):
+    with open(video_path, "rb") as f:
+        data = f.read()
+        b64 = base64.b64encode(data).decode()
+    html = (
+            f"""
+            <video controls autoplay muted loop>
+            <source src="data:video/mp4;base64,{b64}" type="video/mp4">
+            </video> 
+            """
+    )
+    return html
+
 def video_identity(video):
     return video
 
 def nan():
     return None
 
-FORMAT = ['mp4', 'gif'][1]
+FORMAT = ['mp4', 'gif'][0]
 
 def get_huggingface_dataset():
     try:
@@ -78,7 +92,9 @@ def get_huggingface_dataset():
 def update(user_choice, left, right, data_folder=VIDEO_PATH, flag_to_huggingface=True):
     global last_left_video_path 
     global last_right_video_path 
-
+    global last_infer_left_video_path
+    global last_infer_right_video_path
+    
     if flag_to_huggingface: # log
         env_name = str(last_left_video_path).split('/')[1]  # 'robotinder-data/ENV_NAME/'
         current_time = strftime("%Y-%m-%d-%H-%M-%S", gmtime())
@@ -110,7 +126,10 @@ def update(user_choice, left, right, data_folder=VIDEO_PATH, flag_to_huggingface
     right = video_files[selected_video_ids[1]]
     last_left_video_path = left
     last_right_video_path = right
-    return left, right
+    last_infer_left_video_path = inference(left)
+    last_infer_right_video_path = inference(right)
+    
+    return last_infer_left_video_path, last_infer_right_video_path
 
 def replay(left, right):  
     return left, right
@@ -123,6 +142,10 @@ def parse_envs(folder=VIDEO_PATH):
     return envs
 
 def build_interface(iter=3, data_folder=VIDEO_PATH):
+    import sys
+    import csv
+    csv.field_size_limit(sys.maxsize)
+    
     HF_TOKEN = os.getenv('HF_TOKEN')
     print(HF_TOKEN)
     HF_TOKEN = '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'  # my HF token
@@ -130,7 +153,7 @@ def build_interface(iter=3, data_folder=VIDEO_PATH):
     hf_writer = HuggingFaceDatasetSaver(HF_TOKEN, "crowdsourced-robotinder-demo")
     # callback = gr.CSVLogger()
     callback = hf_writer
-    
+
     # build gradio interface
     with gr.Blocks() as demo:
         gr.Markdown("Here is RoboTinder!")
@@ -150,8 +173,13 @@ def build_interface(iter=3, data_folder=VIDEO_PATH):
             left_video_path = video_files[selected_video_ids[0]]
             right_video_path = video_files[selected_video_ids[1]]
             if FORMAT == 'mp4':
-                left = gr.PlayableVideo(left_video_path, label="left_video")
-                right = gr.PlayableVideo(right_video_path, label="right_video")
+                # left = gr.PlayableVideo(left_video_path, label="left_video")
+                # right = gr.PlayableVideo(right_video_path, label="right_video")
+
+                infer_left_video_path = inference(left_video_path)
+                infer_right_video_path = inference(right_video_path)
+                right = gr.HTML(infer_right_video_path, label="right_video")
+                left = gr.HTML(infer_left_video_path, label="left_video")
             else:
                 left = gr.Image(left_video_path, shape=(1024, 768), label="left_video")
                 # right = gr.Image(right_video_path).style(height=768, width=1024)
@@ -162,6 +190,11 @@ def build_interface(iter=3, data_folder=VIDEO_PATH):
         global last_right_video_path 
         last_right_video_path = right_video_path
 
+        global last_infer_left_video_path
+        last_infer_left_video_path = infer_left_video_path
+        global last_infer_right_video_path
+        last_infer_right_video_path = infer_right_video_path
+
         btn1 = gr.Button("Replay")
         user_choice = gr.Radio(["Left", "Right", "Not Sure"], label="Which one is your favorite?")
         btn2 = gr.Button("Next")