summary description and selection

app.py

@@ -36,6 +36,53 @@ def to_string(label):
     return label
 
 
+def summarize_clusters(clusters_list, max_terms=3):
+    for cl_id, cl_dict in enumerate(clusters_list):
+        total = len(cl_dict["img_path_list"])
+        gdr_list = cl_dict["labels_gender"]
+        eth_list = cl_dict["labels_ethnicity"]
+        cl_dict["sentence_desc"] = (
+            f"Cluster {cl_id} | \t"
+            + f"gender terms incl.: {gdr_list[0][0].replace('person', 'unmarked(gender)')}"
+            + (
+                f" - {gdr_list[1][0].replace('person', 'unmarked(gender)')} | "
+                if len(gdr_list) > 1
+                else " | "
+            )
+            + f"ethnicity terms incl.: {'unmarked(ethnicity)' if eth_list[0][0] == '' else eth_list[0][0]}"
+            + (
+                f" - {'unmarked(ethnicity)' if eth_list[1][0] == '' else eth_list[1][0]}"
+                if len(eth_list) > 1
+                else ""
+            )
+        )
+        cl_dict["summary_desc"] = (
+            f"Cluster {cl_id} has {total} images.\n"
+            + f"- The most represented gender terms are {gdr_list[0][0].replace('person', 'unmarked')} ({gdr_list[0][1]})"
+            + (
+                f" and {gdr_list[1][0].replace('person', 'unmarked')} ({gdr_list[1][1]}).\n"
+                if len(gdr_list) > 1
+                else ".\n"
+            )
+            + f"- The most represented ethnicity terms are {'unmarked' if eth_list[0][0] == '' else eth_list[0][0]} ({eth_list[0][1]})"
+            + (
+                f" and {'unmarked' if eth_list[1][0] == '' else eth_list[1][0]} ({eth_list[1][1]}).\n"
+                if len(eth_list) > 1
+                else ".\n"
+            )
+            + "See below for a more detailed description."
+        )
+
+
+for _, clusters_list in clusters_by_size.items():
+    summarize_clusters(clusters_list)
+
+dropdown_descs = dict(
+    (num_clusters, [cl_dct["sentence_desc"] for cl_dct in clusters_list])
+    for num_clusters, clusters_list in clusters_by_size.items()
+)
+
+
 def describe_cluster(cl_dict, block="label", max_items=4):
     labels_values = sorted(cl_dict.items(), key=operator.itemgetter(1))
     labels_values.reverse()
@@ -62,6 +109,12 @@ def describe_cluster(cl_dict, block="label", max_items=4):
 def show_cluster(cl_id, num_clusters):
     if not cl_id:
         cl_id = 0
+    else:
+        cl_id = (
+            dropdown_descs[num_clusters].index(cl_id)
+            if cl_id in dropdown_descs[num_clusters]
+            else 0
+        )
     if not num_clusters:
         num_clusters = 12
     cl_dct = clusters_by_size[num_clusters][cl_id]
@@ -71,9 +124,17 @@ def show_cluster(cl_id, num_clusters):
             [st.replace("/", "") for st in cl_dct["img_path_list"][i].split("//")][3:]
         )
         im = Image.open(os.path.join("identities-images", img_path))
-            #.resize((256, 256))
-        caption = "_".join([img_path.split("/")[0], img_path.split("/")[-1]]).replace("Photo_portrait_of_an_", "").replace("Photo_portrait_of_a_", "").replace("SD_v2_random_seeds_identity_", "(SD v.2) ").replace("dataset-identities-dalle2_", "(Dall-E 2) ").replace("SD_v1.4_random_seeds_identity_", "(SD v.1.4) ").replace("_", " ")             
-        images.append((im,caption))
+        # .resize((256, 256))
+        caption = (
+            "_".join([img_path.split("/")[0], img_path.split("/")[-1]])
+            .replace("Photo_portrait_of_an_", "")
+            .replace("Photo_portrait_of_a_", "")
+            .replace("SD_v2_random_seeds_identity_", "(SD v.2) ")
+            .replace("dataset-identities-dalle2_", "(Dall-E 2) ")
+            .replace("SD_v1.4_random_seeds_identity_", "(SD v.1.4) ")
+            .replace("_", " ")
+        )
+        images.append((im, caption))
     model_fig = go.Figure()
     model_fig.add_trace(
         go.Pie(
@@ -105,7 +166,7 @@ def show_cluster(cl_id, num_clusters):
     )
 
     return (
-        len(cl_dct["img_path_list"]),
+        clusters_by_size[num_clusters][cl_id]["summary_desc"],
         gender_fig,
         gender_description,
         model_fig,
@@ -113,15 +174,18 @@ def show_cluster(cl_id, num_clusters):
         ethnicity_fig,
         ethnicity_description,
         images,
-        gr.update(choices=[i for i in range(num_clusters)]),
+        gr.update(choices=dropdown_descs[num_clusters]),
+        #        gr.update(choices=[i for i in range(num_clusters)]),
     )
 
 
 with gr.Blocks(title=TITLE) as demo:
     gr.Markdown(f"# {TITLE}")
     gr.Markdown(
-        "Explore the data generated from [DiffusionBiasExplorer](https://huggingface.co/spaces/society-ethics/DiffusionBiasExplorer)! This demo showcases patterns in the images generated from different prompts input to Stable Diffusion and Dalle-2 systems.")
-    gr.Markdown("See the results on how the images from different prompts cluster together below."
+        "Explore the data generated from [DiffusionBiasExplorer](https://huggingface.co/spaces/society-ethics/DiffusionBiasExplorer)! This demo showcases patterns in the images generated from different prompts input to Stable Diffusion and Dalle-2 systems."
+    )
+    gr.Markdown(
+        "See the results on how the images from different prompts cluster together below."
     )
     gr.HTML(
         """<span style="color:red" font-size:smaller>⚠️ DISCLAIMER: the images displayed by this tool were generated by text-to-image systems and may depict offensive stereotypes or contain explicit content.</span>"""
@@ -135,13 +199,17 @@ with gr.Blocks(title=TITLE) as demo:
     with gr.Row():
         with gr.Column():
             cluster_id = gr.Dropdown(
-                choices=[i for i in range(num_clusters.value)],
+                choices=dropdown_descs[
+                    num_clusters.value
+                ],  # [i for i in range(num_clusters.value)],
                 value=0,
                 label="Select cluster to visualize:",
             )
-            a = gr.Text(label="Number of images")
+            a = gr.Text(label="Cluster summary")
         with gr.Column():
-            gallery = gr.Gallery(label="Most representative images in cluster").style(grid=[2,4], height="auto")
+            gallery = gr.Gallery(label="Most representative images in cluster").style(
+                grid=[2, 4], height="auto"
+            )
     with gr.Row():
         with gr.Column():
             c = gr.Plot(label="How many images from each system?")
@@ -154,13 +222,10 @@ with gr.Blocks(title=TITLE) as demo:
             d_desc = gr.HTML(label="")
 
     gr.Markdown(
-        f"The 'System makeup' plot corresponds to the number of images from the cluster that come from each of the TTI systems that we are comparing: Dall-E 2, Stable Diffusion v.1.4. and Stable Diffusion v.2."
-    )
-    gr.Markdown(
-        'The Gender plot shows the number of images based on the input prompts that used the words man, woman, non-binary person, and unmarked, which we label "person".'
-    )
-    gr.Markdown(
-        f"The 'Ethnicity label makeup' plot corresponds to the number of images from each of the 18 ethnicities used in the prompts. A blank value means unmarked ethnicity."
+        "### Plot Descriptions \n\n"
+        + " The **System makeup** plot (*left*) corresponds to the number of images from the cluster that come from each of the TTI systems that we are comparing: Dall-E 2, Stable Diffusion v.1.4. and Stable Diffusion v.2.\n\n"
+        + " The **Gender term makeup** plot (*middle*) shows the number of images based on the input prompts that used the phrases man, woman, non-binary person, and person (unmarked) to describe the figure's gender.\n\n"
+        + " The **Ethnicity label makeup** plot (*right*) corresponds to the number of images from each of the 18 ethnicity descriptions used in the prompts. A blank value denotes unmarked ethnicity.\n\n"
     )
     demo.load(
         fn=show_cluster,