feat: Update repack to fit to 512 x 512

DECIMER/Repack_model.py

@@ -5,6 +5,7 @@
 import Transformer_decoder
 import Efficient_Net_encoder
 import config
+import utils
 
 print(tf.__version__)
 
@@ -19,9 +20,9 @@
 max_length = pickle.load(open("max_length_TPU_Stereo.pkl", "rb"))
 
 # Image parameters
-IMG_EMB_DIM = (10, 10, 232)
+IMG_EMB_DIM = (16, 16, 232)
 IMG_EMB_DIM = (IMG_EMB_DIM[0] * IMG_EMB_DIM[1], IMG_EMB_DIM[2])
-IMG_SHAPE = (299, 299, 3)
+IMG_SHAPE = (512, 512, 3)
 PE_INPUT = IMG_EMB_DIM[0]
 IMG_SEQ_LEN, IMG_EMB_DEPTH = IMG_EMB_DIM
 D_MODEL = IMG_EMB_DEPTH
@@ -84,6 +85,17 @@
     start_epoch = int(ckpt_manager.latest_checkpoint.split("-")[-1])
 
 
+def detokenize_output(predicted_array):
+    outputs = [tokenizer.index_word[i] for i in predicted_array[0].numpy()]
+    prediction = (
+        "".join([str(elem) for elem in outputs])
+        .replace("<start>", "")
+        .replace("<end>", "")
+    )
+
+    return prediction
+
+
 class DECIMER_Predictor(tf.Module):
     """This is a class which takes care of inference. It loads the saved checkpoint and the necessary
     tokenizers. The inference begins with the start token (<start>) and ends when the end token(<end>)
@@ -116,10 +128,9 @@ def __call__(self, Decoded_image):
         Returns:
             output (tf.Tensor[tf.int64]): predicted output as an array.
         """
-
         assert isinstance(Decoded_image, tf.Tensor)
         if len(Decoded_image.shape) == 0:
-            Decoded_image = Decoded_image[tf.newaxis]
+            sentence = Decoded_image[tf.newaxis]
 
         _image_batch = tf.expand_dims(Decoded_image, 0)
         _image_embedding = encoder(_image_batch, training=False)