fix: fixing result output and bypassing large files problem

dashboard_image2image.py

@@ -14,7 +14,7 @@ BASELINE_MODEL = "openai/clip-vit-base-patch32"
 MODEL_PATH = "flax-community/clip-rsicd-v2"
 IMAGE_VECTOR_FILE = "./vectors/test-bs128x8-lr5e-6-adam-ckpt-1.tsv"
 IMAGES_DIR = "./images"
-CAPTIONS_FILE = os.path.join(IMAGES_DIR, "dataset_rsicd.json")
+CAPTIONS_FILE = os.path.join(IMAGES_DIR, "test-captions.json")
 
 @st.cache(allow_output_mutation=True)
 def load_example_images():

dashboard_text2image.py

@@ -13,7 +13,7 @@ BASELINE_MODEL = "openai/clip-vit-base-patch32"
 MODEL_PATH = "flax-community/clip-rsicd-v2"
 IMAGE_VECTOR_FILE = "./vectors/test-bs128x8-lr5e-6-adam-ckpt-1.tsv"
 IMAGES_DIR = "./images"
-CAPTIONS_FILE = os.path.join(IMAGES_DIR, "dataset_rsicd.json")
+CAPTIONS_FILE = os.path.join(IMAGES_DIR, "test-captions.json")
 
 def app():
     filenames, index = utils.load_index(IMAGE_VECTOR_FILE)

utils.py

@@ -38,11 +38,9 @@ def load_model(model_path, baseline_model):
 def load_captions(caption_file):
     image2caption = {}
     with open(caption_file, "r") as fcap:
-        data = json.loads(fcap.read())
-        for image in data["images"]:
-            filename = image["filename"]
-            captions = []
-            for sentence in image["sentences"]:
-                captions.append(sentence["raw"])
+        for line in fcap:
+            data = json.loads(line.strip())
+            filename = data["filename"]
+            captions = data["captions"]
             image2caption[filename] = captions
     return image2caption