explore interface

app.py

@@ -22,5 +22,5 @@ demo = gr.TabbedInterface(
     title="Explore & Label Concepts",
     analytics_enabled=True,
 )
-
+      
 demo.launch()

requirements.txt

@@ -0,0 +1 @@
+jsonlines

src/assets/.gitignore

@@ -0,0 +1,2 @@
+*
+!.gitignore

src/constants.py

@@ -0,0 +1,22 @@
+"""Constants use in the space.
+"""
+
+import os
+import pathlib
+
+
+DATASET_NAME = "Xmaster6y/fruit-vegetable-concepts"
+CONCEPTS = [
+    # Shapes
+    "sphere",
+    "cube",
+    "cylinder",
+    # Colors
+    "red",
+    "green",
+    "orange",
+    "yellow",
+]
+ASSETS_FOLDER = pathlib.Path(__file__).parent / "assets"
+
+HF_TOKEN = os.getenv("HF_TOKEN")

src/explore_interface.py

@@ -1,12 +1,70 @@
 """Interface to explore images and labels.
 """
 
+import random
+
 import gradio as gr
 
+from src import global_variables
+from src.constants import CONCEPTS, ASSETS_FOLDER, DATASET_NAME
+
+def get_next_image(
+    split: str,
+    concepts: list,
+    current_metadata: dict,
+    selected_concepts: list,
+):
+    if concepts != selected_concepts:
+        for split, values in global_variables.all_metadata.items():
+            current_metadata[split] = [x for x in values if all([c in x["concepts"] for c in concepts])]
+        selected_concepts = concepts
+    if split == "all":
+        split = random.choice(["train", "validation", "test"])
+    try:
+        sample = random.choice(current_metadata[split])
+        image_path = f"{ASSETS_FOLDER}/{DATASET_NAME}/data/{split}/{sample['file_name']}"
+        return image_path, sample["class"], sample["concepts"], current_metadata, selected_concepts
+    except IndexError:
+        gr.Warning("No image found for the selected filter.")
+        return None, None, None, current_metadata, selected_concepts
 
-with gr.Blocks() as interface:
+with gr.Blocks() as interface:        
     with gr.Row():
         with gr.Column():
-            gr.Image(
+            with gr.Group():
+                gr.Markdown(
+                    "## # Image Selection",
+                )
+                split = gr.Radio(
+                    label="Split",
+                    choices=["train", "validation", "test", "all"],
+                    value="train",
+                )
+                concepts = gr.CheckboxGroup(
+                    label="Concepts",
+                    choices=CONCEPTS,
+                )
+            button = gr.Button(
+                value="Next",
+            )
+            with gr.Group():
+                gr.Markdown(
+                    "##  # Image Info",
+                )
+                im_class = gr.Textbox(
+                    label="Class",
+                )
+                im_concepts = gr.JSON(
+                    label="Concepts",
+                )
+        with gr.Column():
+            image = gr.Image(
                 label="Image",
             )
+    current_metadata = gr.State(global_variables.all_metadata)
+    selected_concepts = gr.State([])
+    button.click(
+        get_next_image, 
+        outputs=[image, im_class, im_concepts, current_metadata, selected_concepts], 
+        inputs=[split, concepts, current_metadata, selected_concepts]
+    )

src/global_variables.py

@@ -0,0 +1,44 @@
+"""Global variables used in the space.
+"""
+
+from huggingface_hub import HfApi
+import jsonlines
+
+import gradio as gr
+
+from src.constants import DATASET_NAME, HF_TOKEN, ASSETS_FOLDER
+
+hf_api: HfApi
+all_metadata: dict
+
+
+def setup():
+    global hf_api
+    global all_metadata
+    hf_api = HfApi(token=HF_TOKEN)
+    hf_api.snapshot_download(
+        local_dir=f"{ASSETS_FOLDER}/{DATASET_NAME}",
+        repo_id=DATASET_NAME,
+        repo_type="dataset",
+    )
+    all_metadata = {}
+    for split in ["train", "validation", "test"]:
+        all_metadata[split] = []
+        with jsonlines.open(f"{ASSETS_FOLDER}/{DATASET_NAME}/data/{split}/metadata.jsonl") as reader:
+            for row in reader:
+                all_metadata[split].append(row)
+
+def save_metadata():
+    global all_metadata
+    for split, values in all_metadata.items():
+        with jsonlines.open(f"{ASSETS_FOLDER}/{DATASET_NAME}/data/{split}/metadata.jsonl", mode='w') as writer:
+            writer.write_all(values)
+        hf_api.upload_file(
+            path_or_fileobj=f"{ASSETS_FOLDER}/{DATASET_NAME}/data/{split}/metadata.jsonl",
+            path_in_repo=f"data/{split}/metadata.jsonl",
+            repo_id=DATASET_NAME,
+            repo_type="dataset",
+        )
+
+if gr.NO_RELOAD:
+    setup()