updates

app.py

@@ -2,9 +2,6 @@ import cv2
 import numpy as np
 import gradio as gr
 
-# import os
-# os.chdir('modeling')
-
 import tensorflow as tf, tf_keras
 import tensorflow_hub as hub
 from transformers import AutoTokenizer, TFAutoModelForSeq2SeqLM
@@ -18,7 +15,7 @@ movinet_model = tf_keras.models.load_model(movinet_path)
 movinet_model.trainable = False
 
 tokenizer = AutoTokenizer.from_pretrained("t5-base")
-t5_model = TFAutoModelForSeq2SeqLM.from_pretrained("deanna-emery/ASL_t5_word_epoch15_1204")
+t5_model = TFAutoModelForSeq2SeqLM.from_pretrained("deanna-emery/ASL_t5_movinet_sentence")
 t5_model.trainable = False
 
 def crop_center_square(frame):
@@ -71,24 +68,34 @@ def translate(video_file):
     
     translation = tokenizer.batch_decode(tokens, skip_special_tokens=True)
 
-    # Return dict {label:pred}
     return {"translation":translation}
 
 # Gradio App config
 title = "ASL Translation (MoViNet + T5)"
 examples = [
-        ['videos/all.mp4', 'all'],
-        ['videos/white.mp4', 'white'],
-        ['videos/before.mp4', 'before'],
-        ['videos/blue.mp4', 'blue'],
+        ["videos/My second ASL professor's name was Will White.mp4", "My second ASL professor's name was Will White"],
+        ["videos/Rainbows rainbows high up in the sky.mp4", "Rainbows rainbows high up in the sky"],
+        ['videos/You are my sunshine.mp4', 'You are my sunshine'],
+        ['videos/scrub your hands for at least 20 seconds.mp4', 'scrub your hands for at least 20 seconds'],
         ['videos/no.mp4', 'no'],
-        ['videos/accident2.mp4', 'accident']
+        ['videos/before.mp4', 'before']
     ]
 
+description =   "Gradio demo of word-level sign language classification using I3D model pretrained on the WLASL video dataset. " \
+                "WLASL is a large-scale dataset containing more than 2000 words in American Sign Language. " \
+                "Examples used in the demo are videos from the the test subset. "  \
+                "Note that WLASL100 contains 100 words while WLASL2000 contains 2000."
+
+
+article =   "More information about the trained models can be found <a href=https://github.com/deanna-emery/ASL-Translator/>here</a>."
+
+
 # Gradio App interface
 gr.Interface(fn=translate,
-              inputs="video", 
+              inputs=[gr.Video(label="Video (*.mp4)"),gr.Radio(label='Caption')], 
               outputs="text",
               allow_flagging="never",
               title=title, 
-              examples=examples).launch()
+              description=description,
+              examples=examples,
+              article=article).launch()