experiment with self-caption

app.py

@@ -40,6 +40,7 @@ model.save_pretrained("./vit-bert")
 model = VisionEncoderDecoderModel.from_pretrained("./vit-bert")
 
 
+'''
 
 repo_name = "ydshieh/vit-gpt2-coco-en"
 test_image = "cats.jpg"
@@ -47,14 +48,16 @@ feature_extractor2 = ViTFeatureExtractor.from_pretrained(repo_name)
 tokenizer = AutoTokenizer.from_pretrained(repo_name)
 model2 = VisionEncoderDecoderModel.from_pretrained(repo_name)
 pixel_values = feature_extractor2(test_image, return_tensors="pt").pixel_values
-
+print("Pixel Values")
+print(pixel_values)
 # autoregressively generate text (using beam search or other decoding strategy)
 generated_ids = model2.generate(pixel_values, max_length=16, num_beams=4, return_dict_in_generate=True)
 # decode into text
 preds = tokenizer.batch_decode(generated_ids[0], skip_special_tokens=True)
 preds = [pred.strip() for pred in preds]
+print("Predictions")
 print(preds)
-'''
+
 
 
 def classify_image(image):