testing object detection with confidence

app.py

@@ -5,7 +5,7 @@ import spaces
 import requests
 import copy
 
-from PIL import Image, ImageDraw, ImageFont 
+from PIL import Image, ImageDraw, ImageFont
 import io
 import matplotlib.pyplot as plt
 import matplotlib.patches as patches
@@ -14,36 +14,88 @@ import random
 import numpy as np
 
 import subprocess
-subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, shell=True)
+
+subprocess.run(
+    "pip install flash-attn --no-build-isolation",
+    env={"FLASH_ATTENTION_SKIP_CUDA_BUILD": "TRUE"},
+    shell=True,
+)
 
 models = {
-    'microsoft/Florence-2-large-ft': AutoModelForCausalLM.from_pretrained('microsoft/Florence-2-large-ft', trust_remote_code=True).to("cuda").eval(),
-    'microsoft/Florence-2-large': AutoModelForCausalLM.from_pretrained('microsoft/Florence-2-large', trust_remote_code=True).to("cuda").eval(),
-    'microsoft/Florence-2-base-ft': AutoModelForCausalLM.from_pretrained('microsoft/Florence-2-base-ft', trust_remote_code=True).to("cuda").eval(),
-    'microsoft/Florence-2-base': AutoModelForCausalLM.from_pretrained('microsoft/Florence-2-base', trust_remote_code=True).to("cuda").eval(),
+    "microsoft/Florence-2-large-ft": AutoModelForCausalLM.from_pretrained(
+        "microsoft/Florence-2-large-ft", trust_remote_code=True
+    )
+    .to("cuda")
+    .eval(),
+    "microsoft/Florence-2-large": AutoModelForCausalLM.from_pretrained(
+        "microsoft/Florence-2-large", trust_remote_code=True
+    )
+    .to("cuda")
+    .eval(),
+    "microsoft/Florence-2-base-ft": AutoModelForCausalLM.from_pretrained(
+        "microsoft/Florence-2-base-ft", trust_remote_code=True
+    )
+    .to("cuda")
+    .eval(),
+    "microsoft/Florence-2-base": AutoModelForCausalLM.from_pretrained(
+        "microsoft/Florence-2-base", trust_remote_code=True
+    )
+    .to("cuda")
+    .eval(),
 }
 
 processors = {
-    'microsoft/Florence-2-large-ft': AutoProcessor.from_pretrained('microsoft/Florence-2-large-ft', trust_remote_code=True),
-    'microsoft/Florence-2-large': AutoProcessor.from_pretrained('microsoft/Florence-2-large', trust_remote_code=True),
-    'microsoft/Florence-2-base-ft': AutoProcessor.from_pretrained('microsoft/Florence-2-base-ft', trust_remote_code=True),
-    'microsoft/Florence-2-base': AutoProcessor.from_pretrained('microsoft/Florence-2-base', trust_remote_code=True),
+    "microsoft/Florence-2-large-ft": AutoProcessor.from_pretrained(
+        "microsoft/Florence-2-large-ft", trust_remote_code=True
+    ),
+    "microsoft/Florence-2-large": AutoProcessor.from_pretrained(
+        "microsoft/Florence-2-large", trust_remote_code=True
+    ),
+    "microsoft/Florence-2-base-ft": AutoProcessor.from_pretrained(
+        "microsoft/Florence-2-base-ft", trust_remote_code=True
+    ),
+    "microsoft/Florence-2-base": AutoProcessor.from_pretrained(
+        "microsoft/Florence-2-base", trust_remote_code=True
+    ),
 }
 
 
 DESCRIPTION = "# [Florence-2 Demo](https://huggingface.co/microsoft/Florence-2-large)"
 
-colormap = ['blue','orange','green','purple','brown','pink','gray','olive','cyan','red',
-            'lime','indigo','violet','aqua','magenta','coral','gold','tan','skyblue']
+colormap = [
+    "blue",
+    "orange",
+    "green",
+    "purple",
+    "brown",
+    "pink",
+    "gray",
+    "olive",
+    "cyan",
+    "red",
+    "lime",
+    "indigo",
+    "violet",
+    "aqua",
+    "magenta",
+    "coral",
+    "gold",
+    "tan",
+    "skyblue",
+]
+
 
 def fig_to_pil(fig):
     buf = io.BytesIO()
-    fig.savefig(buf, format='png')
+    fig.savefig(buf, format="png")
     buf.seek(0)
     return Image.open(buf)
 
+
 @spaces.GPU
-def run_example(task_prompt, image, text_input=None, model_id='microsoft/Florence-2-large'):
+def run_example(
+    task_prompt, image, text_input=None, model_id="microsoft/Florence-2-large"
+):
     model = models[model_id]
     processor = processors[model_id]
     if text_input is None:
@@ -58,38 +110,89 @@ def run_example(task_prompt, image, text_input=None, model_id='microsoft/Florenc
         early_stopping=False,
         do_sample=False,
         num_beams=3,
-        output_scores= True
     )
     generated_text = processor.batch_decode(generated_ids, skip_special_tokens=False)[0]
     parsed_answer = processor.post_process_generation(
-        generated_text,
+        generated_text, task=task_prompt, image_size=(image.width, image.height)
+    )
+    return parsed_answer
+
+
+def run_example_with_score(
+    task_prompt, image, text_input=None, model_id="microsoft/Florence-2-large"
+):
+    model = models[model_id]
+    processor = processors[model_id]
+    if text_input is None:
+        prompt = task_prompt
+    else:
+        prompt = task_prompt + text_input
+    inputs = processor(text=prompt, images=image, return_tensors="pt").to("cuda")
+    generated_ids = model.generate(
+        input_ids=inputs["input_ids"],
+        pixel_values=inputs["pixel_values"],
+        max_new_tokens=1024,
+        num_beams=3,
+        return_dict_in_generate=True,
+        output_scores=True,
+    )
+    generated_text = processor.batch_decode(
+        generated_ids.sequences, skip_special_tokens=False
+    )[0]
+
+    prediction, scores, beam_indices = (
+        generated_ids.sequences,
+        generated_ids.scores,
+        generated_ids.beam_indices,
+    )
+    transition_beam_scores = model.compute_transition_scores(
+        sequences=prediction,
+        scores=scores,
+        beam_indices=beam_indices,
+    )
+
+    parsed_answer = processor.post_process_generation(
+        sequence=generated_ids.sequences[0],
+        transition_beam_score=transition_beam_scores[0],
         task=task_prompt,
-        image_size=(image.width, image.height)
+        image_size=(image.width, image.height),
     )
+
     return parsed_answer
 
+
 def plot_bbox(image, data):
     fig, ax = plt.subplots()
     ax.imshow(image)
-    for bbox, label in zip(data['bboxes'], data['labels']):
+    for bbox, label in zip(data["bboxes"], data["labels"]):
         x1, y1, x2, y2 = bbox
-        rect = patches.Rectangle((x1, y1), x2-x1, y2-y1, linewidth=1, edgecolor='r', facecolor='none')
+        rect = patches.Rectangle(
+            (x1, y1), x2 - x1, y2 - y1, linewidth=1, edgecolor="r", facecolor="none"
+        )
         ax.add_patch(rect)
-        plt.text(x1, y1, label, color='white', fontsize=8, bbox=dict(facecolor='red', alpha=0.5))
-    ax.axis('off')
+        plt.text(
+            x1,
+            y1,
+            label,
+            color="white",
+            fontsize=8,
+            bbox=dict(facecolor="red", alpha=0.5),
+        )
+    ax.axis("off")
     return fig
 
+
 def draw_polygons(image, prediction, fill_mask=False):
 
     draw = ImageDraw.Draw(image)
     scale = 1
-    for polygons, label in zip(prediction['polygons'], prediction['labels']):
+    for polygons, label in zip(prediction["polygons"], prediction["labels"]):
         color = random.choice(colormap)
         fill_color = random.choice(colormap) if fill_mask else None
         for _polygon in polygons:
             _polygon = np.array(_polygon).reshape(-1, 2)
             if len(_polygon) < 3:
-                print('Invalid polygon:', _polygon)
+                print("Invalid polygon:", _polygon)
                 continue
             _polygon = (_polygon * scale).reshape(-1).tolist()
             if fill_mask:
@@ -99,157 +202,176 @@ def draw_polygons(image, prediction, fill_mask=False):
             draw.text((_polygon[0] + 8, _polygon[1] + 2), label, fill=color)
     return image
 
+
 def convert_to_od_format(data):
-    bboxes = data.get('bboxes', [])
-    labels = data.get('bboxes_labels', [])
-    od_results = {
-        'bboxes': bboxes,
-        'labels': labels
-    }
+    bboxes = data.get("bboxes", [])
+    labels = data.get("bboxes_labels", [])
+    od_results = {"bboxes": bboxes, "labels": labels}
     return od_results
 
+
 def draw_ocr_bboxes(image, prediction):
     scale = 1
     draw = ImageDraw.Draw(image)
-    bboxes, labels = prediction['quad_boxes'], prediction['labels']
+    bboxes, labels = prediction["quad_boxes"], prediction["labels"]
     for box, label in zip(bboxes, labels):
         color = random.choice(colormap)
         new_box = (np.array(box) * scale).tolist()
         draw.polygon(new_box, width=3, outline=color)
-        draw.text((new_box[0]+8, new_box[1]+2),
-                  "{}".format(label),
-                  align="right",
-                  fill=color)
+        draw.text(
+            (new_box[0] + 8, new_box[1] + 2),
+            "{}".format(label),
+            align="right",
+            fill=color,
+        )
     return image
 
-def process_image(image, task_prompt, text_input=None, model_id='microsoft/Florence-2-large'):
+
+def process_image(
+    image, task_prompt, text_input=None, model_id="microsoft/Florence-2-large"
+):
     image = Image.fromarray(image)  # Convert NumPy array to PIL Image
-    if task_prompt == 'Caption':
-        task_prompt = '<CAPTION>'
+    if task_prompt == "Caption":
+        task_prompt = "<CAPTION>"
         results = run_example(task_prompt, image, model_id=model_id)
         return results, None
-    elif task_prompt == 'Detailed Caption':
-        task_prompt = '<DETAILED_CAPTION>'
+    elif task_prompt == "Detailed Caption":
+        task_prompt = "<DETAILED_CAPTION>"
         results = run_example(task_prompt, image, model_id=model_id)
         return results, None
-    elif task_prompt == 'More Detailed Caption':
-        task_prompt = '<MORE_DETAILED_CAPTION>'
+    elif task_prompt == "More Detailed Caption":
+        task_prompt = "<MORE_DETAILED_CAPTION>"
         results = run_example(task_prompt, image, model_id=model_id)
         return results, None
-    elif task_prompt == 'Caption + Grounding':
-        task_prompt = '<CAPTION>'
+    elif task_prompt == "Caption + Grounding":
+        task_prompt = "<CAPTION>"
         results = run_example(task_prompt, image, model_id=model_id)
         text_input = results[task_prompt]
-        task_prompt = '<CAPTION_TO_PHRASE_GROUNDING>'
+        task_prompt = "<CAPTION_TO_PHRASE_GROUNDING>"
         results = run_example(task_prompt, image, text_input, model_id)
-        results['<CAPTION>'] = text_input
-        fig = plot_bbox(image, results['<CAPTION_TO_PHRASE_GROUNDING>'])
+        results["<CAPTION>"] = text_input
+        fig = plot_bbox(image, results["<CAPTION_TO_PHRASE_GROUNDING>"])
         return results, fig_to_pil(fig)
-    elif task_prompt == 'Detailed Caption + Grounding':
-        task_prompt = '<DETAILED_CAPTION>'
+    elif task_prompt == "Detailed Caption + Grounding":
+        task_prompt = "<DETAILED_CAPTION>"
         results = run_example(task_prompt, image, model_id=model_id)
         text_input = results[task_prompt]
-        task_prompt = '<CAPTION_TO_PHRASE_GROUNDING>'
+        task_prompt = "<CAPTION_TO_PHRASE_GROUNDING>"
         results = run_example(task_prompt, image, text_input, model_id)
-        results['<DETAILED_CAPTION>'] = text_input
-        fig = plot_bbox(image, results['<CAPTION_TO_PHRASE_GROUNDING>'])
+        results["<DETAILED_CAPTION>"] = text_input
+        fig = plot_bbox(image, results["<CAPTION_TO_PHRASE_GROUNDING>"])
         return results, fig_to_pil(fig)
-    elif task_prompt == 'More Detailed Caption + Grounding':
-        task_prompt = '<MORE_DETAILED_CAPTION>'
+    elif task_prompt == "More Detailed Caption + Grounding":
+        task_prompt = "<MORE_DETAILED_CAPTION>"
         results = run_example(task_prompt, image, model_id=model_id)
         text_input = results[task_prompt]
-        task_prompt = '<CAPTION_TO_PHRASE_GROUNDING>'
+        task_prompt = "<CAPTION_TO_PHRASE_GROUNDING>"
         results = run_example(task_prompt, image, text_input, model_id)
-        results['<MORE_DETAILED_CAPTION>'] = text_input
-        fig = plot_bbox(image, results['<CAPTION_TO_PHRASE_GROUNDING>'])
+        results["<MORE_DETAILED_CAPTION>"] = text_input
+        fig = plot_bbox(image, results["<CAPTION_TO_PHRASE_GROUNDING>"])
         return results, fig_to_pil(fig)
-    elif task_prompt == 'Object Detection':
-        task_prompt = '<OD>'
-        results = run_example(task_prompt, image, model_id=model_id)
-        fig = plot_bbox(image, results['<OD>'])
+    elif task_prompt == "Object Detection":
+        task_prompt = "<OD>"
+        results = run_example_with_score(task_prompt, image, model_id=model_id)
+        fig = plot_bbox(image, results["<OD>"])
         return results, fig_to_pil(fig)
-    elif task_prompt == 'Dense Region Caption':
-        task_prompt = '<DENSE_REGION_CAPTION>'
+    elif task_prompt == "Dense Region Caption":
+        task_prompt = "<DENSE_REGION_CAPTION>"
         results = run_example(task_prompt, image, model_id=model_id)
-        fig = plot_bbox(image, results['<DENSE_REGION_CAPTION>'])
+        fig = plot_bbox(image, results["<DENSE_REGION_CAPTION>"])
         return results, fig_to_pil(fig)
-    elif task_prompt == 'Region Proposal':
-        task_prompt = '<REGION_PROPOSAL>'
+    elif task_prompt == "Region Proposal":
+        task_prompt = "<REGION_PROPOSAL>"
         results = run_example(task_prompt, image, model_id=model_id)
-        fig = plot_bbox(image, results['<REGION_PROPOSAL>'])
+        fig = plot_bbox(image, results["<REGION_PROPOSAL>"])
         return results, fig_to_pil(fig)
-    elif task_prompt == 'Caption to Phrase Grounding':
-        task_prompt = '<CAPTION_TO_PHRASE_GROUNDING>'
+    elif task_prompt == "Caption to Phrase Grounding":
+        task_prompt = "<CAPTION_TO_PHRASE_GROUNDING>"
         results = run_example(task_prompt, image, text_input, model_id)
-        fig = plot_bbox(image, results['<CAPTION_TO_PHRASE_GROUNDING>'])
+        fig = plot_bbox(image, results["<CAPTION_TO_PHRASE_GROUNDING>"])
         return results, fig_to_pil(fig)
-    elif task_prompt == 'Referring Expression Segmentation':
-        task_prompt = '<REFERRING_EXPRESSION_SEGMENTATION>'
+    elif task_prompt == "Referring Expression Segmentation":
+        task_prompt = "<REFERRING_EXPRESSION_SEGMENTATION>"
         results = run_example(task_prompt, image, text_input, model_id)
         output_image = copy.deepcopy(image)
-        output_image = draw_polygons(output_image, results['<REFERRING_EXPRESSION_SEGMENTATION>'], fill_mask=True)
+        output_image = draw_polygons(
+            output_image, results["<REFERRING_EXPRESSION_SEGMENTATION>"], fill_mask=True
+        )
         return results, output_image
-    elif task_prompt == 'Region to Segmentation':
-        task_prompt = '<REGION_TO_SEGMENTATION>'
+    elif task_prompt == "Region to Segmentation":
+        task_prompt = "<REGION_TO_SEGMENTATION>"
         results = run_example(task_prompt, image, text_input, model_id)
         output_image = copy.deepcopy(image)
-        output_image = draw_polygons(output_image, results['<REGION_TO_SEGMENTATION>'], fill_mask=True)
+        output_image = draw_polygons(
+            output_image, results["<REGION_TO_SEGMENTATION>"], fill_mask=True
+        )
         return results, output_image
-    elif task_prompt == 'Open Vocabulary Detection':
-        task_prompt = '<OPEN_VOCABULARY_DETECTION>'
+    elif task_prompt == "Open Vocabulary Detection":
+        task_prompt = "<OPEN_VOCABULARY_DETECTION>"
         results = run_example(task_prompt, image, text_input, model_id)
-        bbox_results = convert_to_od_format(results['<OPEN_VOCABULARY_DETECTION>'])
+        bbox_results = convert_to_od_format(results["<OPEN_VOCABULARY_DETECTION>"])
         fig = plot_bbox(image, bbox_results)
         return results, fig_to_pil(fig)
-    elif task_prompt == 'Region to Category':
-        task_prompt = '<REGION_TO_CATEGORY>'
+    elif task_prompt == "Region to Category":
+        task_prompt = "<REGION_TO_CATEGORY>"
         results = run_example(task_prompt, image, text_input, model_id)
         return results, None
-    elif task_prompt == 'Region to Description':
-        task_prompt = '<REGION_TO_DESCRIPTION>'
+    elif task_prompt == "Region to Description":
+        task_prompt = "<REGION_TO_DESCRIPTION>"
         results = run_example(task_prompt, image, text_input, model_id)
         return results, None
-    elif task_prompt == 'OCR':
-        task_prompt = '<OCR>'
+    elif task_prompt == "OCR":
+        task_prompt = "<OCR>"
         results = run_example(task_prompt, image, model_id=model_id)
         return results, None
-    elif task_prompt == 'OCR with Region':
-        task_prompt = '<OCR_WITH_REGION>'
+    elif task_prompt == "OCR with Region":
+        task_prompt = "<OCR_WITH_REGION>"
         results = run_example(task_prompt, image, model_id=model_id)
         output_image = copy.deepcopy(image)
-        output_image = draw_ocr_bboxes(output_image, results['<OCR_WITH_REGION>'])
+        output_image = draw_ocr_bboxes(output_image, results["<OCR_WITH_REGION>"])
         return results, output_image
     else:
         return "", None  # Return empty string and None for unknown task prompts
 
+
 css = """
   #output {
-    height: 500px; 
-    overflow: auto; 
-    border: 1px solid #ccc; 
+    height: 500px;
+    overflow: auto;
+    border: 1px solid #ccc;
   }
 """
 
 
-single_task_list =[
-    'Caption', 'Detailed Caption', 'More Detailed Caption', 'Object Detection',
-    'Dense Region Caption', 'Region Proposal', 'Caption to Phrase Grounding',
-    'Referring Expression Segmentation', 'Region to Segmentation',
-    'Open Vocabulary Detection', 'Region to Category', 'Region to Description',
-    'OCR', 'OCR with Region'
+single_task_list = [
+    "Caption",
+    "Detailed Caption",
+    "More Detailed Caption",
+    "Object Detection",
+    "Dense Region Caption",
+    "Region Proposal",
+    "Caption to Phrase Grounding",
+    "Referring Expression Segmentation",
+    "Region to Segmentation",
+    "Open Vocabulary Detection",
+    "Region to Category",
+    "Region to Description",
+    "OCR",
+    "OCR with Region",
 ]
 
-cascased_task_list =[
-    'Caption + Grounding', 'Detailed Caption + Grounding', 'More Detailed Caption + Grounding'
+cascased_task_list = [
+    "Caption + Grounding",
+    "Detailed Caption + Grounding",
+    "More Detailed Caption + Grounding",
 ]
 
 
 def update_task_dropdown(choice):
-    if choice == 'Cascased task':
-        return gr.Dropdown(choices=cascased_task_list, value='Caption + Grounding')
+    if choice == "Cascased task":
+        return gr.Dropdown(choices=cascased_task_list, value="Caption + Grounding")
     else:
-        return gr.Dropdown(choices=single_task_list, value='Caption')
-
+        return gr.Dropdown(choices=single_task_list, value="Caption")
 
 
 with gr.Blocks(css=css) as demo:
@@ -258,10 +380,22 @@ with gr.Blocks(css=css) as demo:
         with gr.Row():
             with gr.Column():
                 input_img = gr.Image(label="Input Picture")
-                model_selector = gr.Dropdown(choices=list(models.keys()), label="Model", value='microsoft/Florence-2-large')
-                task_type = gr.Radio(choices=['Single task', 'Cascased task'], label='Task type selector', value='Single task')
-                task_prompt = gr.Dropdown(choices=single_task_list, label="Task Prompt", value="Caption")
-                task_type.change(fn=update_task_dropdown, inputs=task_type, outputs=task_prompt)
+                model_selector = gr.Dropdown(
+                    choices=list(models.keys()),
+                    label="Model",
+                    value="microsoft/Florence-2-large",
+                )
+                task_type = gr.Radio(
+                    choices=["Single task", "Cascased task"],
+                    label="Task type selector",
+                    value="Single task",
+                )
+                task_prompt = gr.Dropdown(
+                    choices=single_task_list, label="Task Prompt", value="Caption"
+                )
+                task_type.change(
+                    fn=update_task_dropdown, inputs=task_type, outputs=task_prompt
+                )
                 text_input = gr.Textbox(label="Text Input (optional)")
                 submit_btn = gr.Button(value="Submit")
             with gr.Column():
@@ -270,16 +404,20 @@ with gr.Blocks(css=css) as demo:
 
         gr.Examples(
             examples=[
-                ["image1.jpg", 'Object Detection'],
-                ["image2.jpg", 'OCR with Region']
+                ["image1.jpg", "Object Detection"],
+                ["image2.jpg", "OCR with Region"],
             ],
             inputs=[input_img, task_prompt],
             outputs=[output_text, output_img],
             fn=process_image,
             cache_examples=True,
-            label='Try examples'
+            label="Try examples",
         )
 
-        submit_btn.click(process_image, [input_img, task_prompt, text_input, model_selector], [output_text, output_img])
+        submit_btn.click(
+            process_image,
+            [input_img, task_prompt, text_input, model_selector],
+            [output_text, output_img],
+        )
 
 demo.launch(debug=True)