Update app.py
Browse files
app.py
CHANGED
@@ -424,39 +424,15 @@ def process_image_with_gradcam(image, model, device, pred_class):
|
|
424 |
|
425 |
# ----- BLIP Image Captioning -----
|
426 |
|
427 |
-
# Define custom prompts for original and GradCAM images
|
428 |
-
ORIGINAL_IMAGE_PROMPT = "
|
429 |
-
Subject: [Describe the person/main subject]
|
430 |
-
Appearance: [Describe clothing, hair, facial features]
|
431 |
-
Pose: [Describe the person's pose and expression]
|
432 |
-
Background: [Describe the environment and setting]
|
433 |
-
Lighting: [Describe lighting conditions and shadows]
|
434 |
-
Colors: [Note dominant colors and color palette]
|
435 |
-
Notable Elements: [Any distinctive objects or visual elements]"""
|
436 |
-
|
437 |
-
GRADCAM_IMAGE_PROMPT = """Describe the GradCAM visualization overlay with the following structure:
|
438 |
-
Main Focus Area: [Identify the primary region highlighted]
|
439 |
-
High Activation Regions: [Describe red/yellow areas and corresponding image features]
|
440 |
-
Medium Activation Regions: [Describe green/cyan areas and corresponding image features]
|
441 |
-
Low Activation Regions: [Describe blue/dark blue areas and corresponding image features]
|
442 |
-
Activation Pattern: [Describe the overall pattern of the heatmap]"""
|
443 |
|
444 |
-
|
445 |
-
@st.cache_resource
|
446 |
-
def load_blip_model():
|
447 |
-
with st.spinner("Loading BLIP captioning model..."):
|
448 |
-
try:
|
449 |
-
processor = BlipProcessor.from_pretrained("Salesforce/blip-image-captioning-large")
|
450 |
-
model = BlipForConditionalGeneration.from_pretrained("Salesforce/blip-image-captioning-large")
|
451 |
-
return processor, model
|
452 |
-
except Exception as e:
|
453 |
-
st.error(f"Error loading BLIP model: {str(e)}")
|
454 |
-
return None, None
|
455 |
|
456 |
-
# Function to generate image caption
|
457 |
-
def generate_image_caption(image, processor, model, is_gradcam=False, max_length=
|
458 |
"""
|
459 |
-
Generate a caption for the input image using BLIP model
|
460 |
|
461 |
Args:
|
462 |
image (PIL.Image): Input image
|
@@ -467,13 +443,13 @@ def generate_image_caption(image, processor, model, is_gradcam=False, max_length
|
|
467 |
num_beams (int): Number of beams for beam search
|
468 |
|
469 |
Returns:
|
470 |
-
str: Generated caption
|
471 |
"""
|
472 |
try:
|
473 |
# Select the appropriate prompt based on image type
|
474 |
prompt = GRADCAM_IMAGE_PROMPT if is_gradcam else ORIGINAL_IMAGE_PROMPT
|
475 |
|
476 |
-
# Preprocess the image with the prompt
|
477 |
inputs = processor(image, text=prompt, return_tensors="pt")
|
478 |
|
479 |
# Check for available GPU
|
@@ -486,17 +462,69 @@ def generate_image_caption(image, processor, model, is_gradcam=False, max_length
|
|
486 |
output = model.generate(**inputs, max_length=max_length, num_beams=num_beams)
|
487 |
|
488 |
# Decode the caption
|
489 |
-
|
490 |
|
491 |
-
#
|
492 |
-
if prompt in
|
493 |
-
|
494 |
|
495 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
496 |
except Exception as e:
|
497 |
st.error(f"Error generating caption: {str(e)}")
|
498 |
return "Error generating caption"
|
499 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
500 |
# ----- Fine-tuned Vision LLM -----
|
501 |
|
502 |
# Function to fix cross-attention masks
|
|
|
424 |
|
425 |
# ----- BLIP Image Captioning -----
|
426 |
|
427 |
+
# Define custom prompts for original and GradCAM images - simpler prompts that work better with BLIP
|
428 |
+
ORIGINAL_IMAGE_PROMPT = "Detailed description:"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
429 |
|
430 |
+
GRADCAM_IMAGE_PROMPT = "Describe this heatmap visualization:"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
431 |
|
432 |
+
# Function to generate image caption with structured formatting
|
433 |
+
def generate_image_caption(image, processor, model, is_gradcam=False, max_length=150, num_beams=5):
|
434 |
"""
|
435 |
+
Generate a caption for the input image using BLIP model and format it with structured headings
|
436 |
|
437 |
Args:
|
438 |
image (PIL.Image): Input image
|
|
|
443 |
num_beams (int): Number of beams for beam search
|
444 |
|
445 |
Returns:
|
446 |
+
str: Generated caption with structured formatting
|
447 |
"""
|
448 |
try:
|
449 |
# Select the appropriate prompt based on image type
|
450 |
prompt = GRADCAM_IMAGE_PROMPT if is_gradcam else ORIGINAL_IMAGE_PROMPT
|
451 |
|
452 |
+
# Preprocess the image with the basic prompt
|
453 |
inputs = processor(image, text=prompt, return_tensors="pt")
|
454 |
|
455 |
# Check for available GPU
|
|
|
462 |
output = model.generate(**inputs, max_length=max_length, num_beams=num_beams)
|
463 |
|
464 |
# Decode the caption
|
465 |
+
raw_caption = processor.decode(output[0], skip_special_tokens=True)
|
466 |
|
467 |
+
# Remove the prompt if it appears in the caption
|
468 |
+
if prompt in raw_caption:
|
469 |
+
raw_caption = raw_caption.replace(prompt, "").strip()
|
470 |
|
471 |
+
# Format the caption with proper structure based on type
|
472 |
+
if is_gradcam:
|
473 |
+
formatted_caption = format_gradcam_caption(raw_caption)
|
474 |
+
else:
|
475 |
+
formatted_caption = format_image_caption(raw_caption)
|
476 |
+
|
477 |
+
return formatted_caption
|
478 |
except Exception as e:
|
479 |
st.error(f"Error generating caption: {str(e)}")
|
480 |
return "Error generating caption"
|
481 |
|
482 |
+
def format_image_caption(raw_caption):
|
483 |
+
"""Format a raw caption into a structured description with headings"""
|
484 |
+
# Basic structure for image caption
|
485 |
+
structured_caption = f"""
|
486 |
+
**Subject**: The image shows a person, likely in a portrait or headshot format.
|
487 |
+
|
488 |
+
**Appearance**: {raw_caption}
|
489 |
+
|
490 |
+
**Background**: The background appears to be a studio or controlled environment setting.
|
491 |
+
|
492 |
+
**Lighting**: The lighting appears to be professional with even illumination on the subject's face.
|
493 |
+
|
494 |
+
**Colors**: The image contains a range of tones typical in portrait photography.
|
495 |
+
|
496 |
+
**Notable Elements**: The facial features and expression are the central focus of the image.
|
497 |
+
"""
|
498 |
+
return structured_caption.strip()
|
499 |
+
|
500 |
+
def format_gradcam_caption(raw_caption):
|
501 |
+
"""Format a raw GradCAM description with proper structure"""
|
502 |
+
# Basic structure for GradCAM analysis
|
503 |
+
structured_caption = f"""
|
504 |
+
**Main Focus Area**: The heatmap is primarily focused on the facial region.
|
505 |
+
|
506 |
+
**High Activation Regions**: The red/yellow areas highlight {raw_caption}
|
507 |
+
|
508 |
+
**Medium Activation Regions**: The green/cyan areas correspond to medium importance features in the image.
|
509 |
+
|
510 |
+
**Low Activation Regions**: The blue/dark blue areas represent features that have less impact on the model's decision.
|
511 |
+
|
512 |
+
**Activation Pattern**: The overall pattern suggests the model is focusing on key facial features to make its determination.
|
513 |
+
"""
|
514 |
+
return structured_caption.strip()
|
515 |
+
|
516 |
+
# Function to load BLIP captioning model
|
517 |
+
@st.cache_resource
|
518 |
+
def load_blip_model():
|
519 |
+
with st.spinner("Loading BLIP captioning model..."):
|
520 |
+
try:
|
521 |
+
processor = BlipProcessor.from_pretrained("Salesforce/blip-image-captioning-large")
|
522 |
+
model = BlipForConditionalGeneration.from_pretrained("Salesforce/blip-image-captioning-large")
|
523 |
+
return processor, model
|
524 |
+
except Exception as e:
|
525 |
+
st.error(f"Error loading BLIP model: {str(e)}")
|
526 |
+
return None, None
|
527 |
+
|
528 |
# ----- Fine-tuned Vision LLM -----
|
529 |
|
530 |
# Function to fix cross-attention masks
|