mammal collage for Field Museum

app.py

@@ -468,11 +468,11 @@ def content_input_images(col_left, col_right):
             pass
     
     with col_left:
-        # if st.session_state.is_hf:
-        st.session_state['dir_uploaded_images'] = os.path.join(st.session_state.dir_home,'uploads')
-        st.session_state['dir_uploaded_images_small'] = os.path.join(st.session_state.dir_home,'uploads_small')
-        uploaded_files = st.file_uploader("Upload Images", type=['jpg', 'jpeg','pdf'], accept_multiple_files=True, key=st.session_state['uploader_idk'])
-        st.button("Use Test Image",help="This will clear any uploaded images and load the 1 provided test image.",on_click=use_test_image)
+        if st.session_state.is_hf:
+            st.session_state['dir_uploaded_images'] = os.path.join(st.session_state.dir_home,'uploads')
+            st.session_state['dir_uploaded_images_small'] = os.path.join(st.session_state.dir_home,'uploads_small')
+            uploaded_files = st.file_uploader("Upload Images", type=['jpg', 'jpeg','pdf'], accept_multiple_files=True, key=st.session_state['uploader_idk'])
+            st.button("Use Test Image",help="This will clear any uploaded images and load the 1 provided test image.",on_click=use_test_image)
     
     with col_right:
         if st.session_state.is_hf:
@@ -2218,11 +2218,52 @@ def content_collage_overlay():
         st.markdown("Prior to transcription, use LeafMachine2 to crop all labels from input images to create label collages for each specimen image. Showing just the text labels to the OCR algorithms significantly improves performance. This runs slowly on the free Hugging Face Space, but runs quickly with a fast CPU or any GPU.")
         st.markdown("Images that are mostly text (like a scanned notecard, or already cropped images) do not require LM2 collage.")
 
-        if st.session_state.is_hf:
-            st.session_state.config['leafmachine']['use_RGB_label_images'] = st.checkbox(":rainbow[Use LeafMachine2 label collage for transcriptions]", st.session_state.config['leafmachine'].get('use_RGB_label_images', False), key='do make collage hf')
-        else:
-            st.session_state.config['leafmachine']['use_RGB_label_images'] = st.checkbox(":rainbow[Use LeafMachine2 label collage for transcriptions]", st.session_state.config['leafmachine'].get('use_RGB_label_images', True), key='do make collage local')
+        # if st.session_state.is_hf:
+        #     st.session_state.config['leafmachine']['use_RGB_label_images'] = st.checkbox(":rainbow[Use LeafMachine2 label collage for transcriptions]", st.session_state.config['leafmachine'].get('use_RGB_label_images', False), key='do make collage hf')
+        # else:
+        #     st.session_state.config['leafmachine']['use_RGB_label_images'] = st.checkbox(":rainbow[Use LeafMachine2 label collage for transcriptions]", st.session_state.config['leafmachine'].get('use_RGB_label_images', True), key='do make collage local')
+        # Set the options for the radio button
+        # Set the options for the radio button with corresponding indices
+        # Set the options for the transcription method radio button
+        options = {
+            0: "Use LeafMachine2 label collage for transcriptions",
+            1: "Use original images for transcriptions",
+            2: "Use specimen collage for transcriptions"
+        }
 
+        # Determine the default index based on the current configuration
+        default_index = st.session_state.config['leafmachine'].get('use_RGB_label_images', 0)
+
+        # Create the radio button for transcription method selection
+        selected_option = st.radio(
+            "Select the transcription method:",
+            options=list(options.values()),
+            index=default_index
+        )
+
+        # Update the session state based on the selected option
+        selected_index = list(options.values()).index(selected_option)
+        st.session_state.config['leafmachine']['use_RGB_label_images'] = selected_index
+
+        # If "Use specimen collage for transcriptions" is selected, show another radio button for rotation options
+        if selected_index == 2:
+            rotation_options = {
+                True: "Rotate clockwise",
+                False: "Rotate counterclockwise"
+            }
+            
+            # Determine the default rotation direction
+            default_rotation = st.session_state.config['leafmachine']['project'].get('specimen_rotate', True)
+            
+            # Create the radio button for rotation direction selection
+            selected_rotation = st.radio(
+                "Select the rotation direction:",
+                options=list(rotation_options.values()),
+                index=0 if default_rotation else 1
+            )
+            
+            # Update the configuration based on the selected rotation direction
+            st.session_state.config['leafmachine']['project']['specimen_rotate'] = selected_rotation == "Rotate clockwise"
 
         option_selected_crops = st.multiselect(label="Components to crop",  
                 options=['ruler', 'barcode','label', 'colorcard','map','envelope','photo','attached_item','weights',

vouchervision/VoucherVision_Config_Builder.py

@@ -53,7 +53,7 @@ def build_VV_config(loaded_cfg=None):
 
         LLM_version_user = 'Gemini 1.5 Flash' # 'Azure GPT 4' #'Azure GPT 4 Turbo 1106-preview'
         prompt_version = 'SLTPvM_long.yaml' # from ["Version 1", "Version 1 No Domain Knowledge", "Version 2"]
-        use_LeafMachine2_collage_images = True # Use LeafMachine2 collage images
+        use_LeafMachine2_collage_images = 0 # Use LeafMachine2 collage images [0, 1, 2]
         do_create_OCR_helper_image = True
 
         batch_size = 500
@@ -65,6 +65,8 @@ def build_VV_config(loaded_cfg=None):
         path_domain_knowledge = '' #os.path.join(dir_home,'domain_knowledge','SLTP_UM_AllAsiaMinimalInRegion.xlsx')
         embeddings_database_name = '' #os.path.splitext(os.path.basename(path_domain_knowledge))[0]
 
+        specimen_rotate = False # False = counter clock, True = clock
+
         #############################################
         #############################################
         ########## DO NOT EDIT BELOW HERE ###########
@@ -75,7 +77,7 @@ def build_VV_config(loaded_cfg=None):
                         path_domain_knowledge,embeddings_database_name,use_LeafMachine2_collage_images,
                         prompt_version, do_create_OCR_helper_image, do_use_trOCR, do_use_florence, trOCR_model_path, florence_model_path, OCR_option, OCR_option_llava, 
                         OCR_option_llava_bit, OCR_GPT_4o_mini_resolution, double_OCR, save_cropped_annotations, 
-                        tool_GEO, tool_WFO, tool_wikipedia,
+                        tool_GEO, tool_WFO, tool_wikipedia,specimen_rotate,
                         check_for_illegal_filenames, skip_vertical, pdf_conversion_dpi, use_domain_knowledge=False)
     else:
         dir_home = os.path.dirname(os.path.dirname(__file__))
@@ -103,6 +105,8 @@ def build_VV_config(loaded_cfg=None):
         tool_WFO = loaded_cfg['leafmachine']['project']['tool_WFO']
         tool_wikipedia = loaded_cfg['leafmachine']['project']['tool_wikipedia']
 
+        specimen_rotate = loaded_cfg['leafmachine']['project']['specimen_rotate']
+
         pdf_conversion_dpi = loaded_cfg['leafmachine']['project']['pdf_conversion_dpi']
 
         LLM_version_user = loaded_cfg['leafmachine']['LLM_version']
@@ -125,7 +129,7 @@ def build_VV_config(loaded_cfg=None):
                         path_domain_knowledge,embeddings_database_name,use_LeafMachine2_collage_images,
                         prompt_version, do_create_OCR_helper_image, do_use_trOCR, do_use_florence, trOCR_model_path, florence_model_path, OCR_option, OCR_option_llava, 
                         OCR_option_llava_bit, OCR_GPT_4o_mini_resolution, double_OCR, save_cropped_annotations,
-                        tool_GEO, tool_WFO, tool_wikipedia,
+                        tool_GEO, tool_WFO, tool_wikipedia,specimen_rotate,
                         check_for_illegal_filenames, skip_vertical, pdf_conversion_dpi, use_domain_knowledge=False)
 
 
@@ -134,7 +138,7 @@ def assemble_config(dir_home, run_name, dir_images_local,dir_output,
                     path_domain_knowledge,embeddings_database_name,use_LeafMachine2_collage_images,
                     prompt_version, do_create_OCR_helper_image_user, do_use_trOCR, do_use_florence, trOCR_model_path, florence_model_path, OCR_option, OCR_option_llava, 
                     OCR_option_llava_bit, OCR_GPT_4o_mini_resolution, double_OCR, save_cropped_annotations, 
-                    tool_GEO, tool_WFO, tool_wikipedia,
+                    tool_GEO, tool_WFO, tool_wikipedia,specimen_rotate,
                     check_for_illegal_filenames, skip_vertical, pdf_conversion_dpi, use_domain_knowledge=False):
     
 
@@ -191,6 +195,7 @@ def assemble_config(dir_home, run_name, dir_images_local,dir_output,
         'tool_GEO': tool_GEO, 
         'tool_WFO': tool_WFO, 
         'tool_wikipedia': tool_wikipedia,
+        'specimen_rotate': specimen_rotate,
     }
 
     modules_section = {

vouchervision/VoucherVision_GUI.py

@@ -1576,8 +1576,22 @@ def content_collage_overlay():
     with col_cropped_1:
         default_crops = st.session_state.config['leafmachine']['cropped_components']['save_cropped_annotations']
         st.write("Prior to transcription, use LeafMachine2 to crop all labels from input images to create label collages for each specimen image. (Requires GPU)")
-        st.session_state.config['leafmachine']['use_RGB_label_images'] = st.checkbox("Use LeafMachine2 label collage for transcriptions", st.session_state.config['leafmachine'].get('use_RGB_label_images', False))
+        # Set the options for the radio button
+        options = {
+            "Use LeafMachine2 label collage for transcriptions": "use_RGB_label_images",
+            "Use specimen collage for transcriptions": "use_specimen_collage"
+        }
+
+        # Create the radio button with the available options
+        selected_option = st.radio(
+            "Select the transcription method:",
+            options=list(options.keys()),
+            index=0 if st.session_state.config['leafmachine'].get('use_RGB_label_images', False) else 1
+        )
 
+        # Update the session state based on the selected option
+        st.session_state.config['leafmachine']['use_RGB_label_images'] = (selected_option == "Use LeafMachine2 label collage for transcriptions")
+        st.session_state.config['leafmachine']['project']['use_specimen_collage'] = (selected_option == "Use specimen collage for transcriptions")
 
         option_selected_crops = st.multiselect(label="Components to crop",  
                 options=['ruler', 'barcode','label', 'colorcard','map','envelope','photo','attached_item','weights',

vouchervision/data_project.py

@@ -116,14 +116,24 @@ class Project_Info():
         n_total = len(os.listdir(dir_images))
         for file in tqdm(os.listdir(dir_images), desc=f'{bcolors.HEADER}     Copying images to working directory{bcolors.ENDC}', colour="white", position=0, total=n_total):
             source = os.path.join(dir_images, file)
+            
+            # Split the filename and the extension
+            filename, ext = os.path.splitext(file)
+            
+            # Convert the extension to lower case
+            ext_lower = ext.lower()
+            
+            # Reconstruct the file name with the lower-cased extension
+            file_with_lower_ext = f"{filename}{ext_lower}"
+
             # Check if file is a PDF
-            if file.lower().endswith('.pdf'):
+            if ext_lower == '.pdf':
                 # Convert PDF pages to JPG images
                 n_pages = convert_pdf_to_jpg(source, Dirs.save_original)
                 self.logger.info(f"Converted {n_pages} pages to JPG from PDF: {file}")
             else:
-                # Copy non-PDF files directly
-                destination = os.path.join(Dirs.save_original, file)
+                # Copy non-PDF files directly, using the filename with the lower-cased extension
+                destination = os.path.join(Dirs.save_original, file_with_lower_ext)
                 shutil.copy(source, destination)
         
     def make_file_names_custom(self, dir_images, cfg, Dirs): 

vouchervision/general_utils.py

@@ -9,6 +9,7 @@ import numpy as np
 import concurrent.futures
 from time import perf_counter
 import torch
+from collections import defaultdict
 
 try:
     from vouchervision.model_maps import ModelMaps
@@ -928,6 +929,7 @@ def crop_detections_from_images(cfg, logger, dir_home, Project, Dirs, batch_size
 def crop_detections_from_images_VV(cfg, logger, dir_home, Project, Dirs, batch_size=50):
     t2_start = perf_counter()
     logger.name = 'Crop Components'
+
     
     if cfg['leafmachine']['cropped_components']['do_save_cropped_annotations']:
         detections = cfg['leafmachine']['cropped_components']['save_cropped_annotations']
@@ -1152,6 +1154,7 @@ def crop_component_from_yolo_coords_VV(anno_type, Dirs, analysis, all_detections
                 # print(f'detection_class: {detection_class} not in save_list: {save_list}')
                 pass
 
+    ### Below creates the LM2 Label Collage image
     # Initialize a list to hold all the acceptable cropped images
     acceptable_cropped_images = []
 
@@ -1243,7 +1246,68 @@ def crop_component_from_yolo_coords_VV(anno_type, Dirs, analysis, all_detections
         original_image_name = '.'.join([filename,'jpg'])
         cv2.imwrite(os.path.join(Dirs.save_original, original_image_name), full_image)
         
+def create_specimen_collage(cfg, logger, dir_home, Project, Dirs):
+    if cfg['leafmachine']['use_RGB_label_images'] == 2:
+        # Get all filenames in the save_original directory that end with .jpg or .jpeg
+        filenames = [f for f in os.listdir(Dirs.save_original) if f.lower().endswith(('.jpg', '.jpeg'))]
+
+        # Dictionary to group filenames by their file stem (e.g., FMNH_6238)
+        grouped_filenames = defaultdict(list)
+
+        for filename in filenames:
+            parts = filename.rsplit('_', 1)
+            if len(parts) == 2 and parts[1][0].isalnum():
+                file_stem = parts[0]
+                grouped_filenames[file_stem].append(filename)
+            else:
+                logger.warning(f"Filename {filename} does not match expected pattern. Skipping.")
+
+        # Process each group of images
+        for file_stem, group in grouped_filenames.items():
+            # Load all cropped images for the current group
+            cropped_images = [cv2.imread(os.path.join(Dirs.save_original, filename)) for filename in group]
+
+            if not cropped_images:
+                logger.error(f"No images found for {file_stem}. Skipping collage creation.")
+                continue
+
+            # Rotate images so that width is greater than height
+            for i, img in enumerate(cropped_images):
+                if img.shape[0] > img.shape[1]:  # height > width
+                    if cfg['leafmachine']['project']['specimen_rotate']:
+                        cropped_images[i] = cv2.rotate(img, cv2.ROTATE_90_CLOCKWISE)
+                    else:
+                        cropped_images[i] = cv2.rotate(img, cv2.ROTATE_90_COUNTERCLOCKWISE)
+
+            # Calculate the maximum width and total height required for the collage
+            max_width = max(img.shape[1] for img in cropped_images)
+            total_height = sum(img.shape[0] for img in cropped_images)
+
+            # Create a black image with the required dimensions
+            collage_image = np.zeros((total_height, max_width, 3), dtype=np.uint8)
 
+            # Stack images on top of each other
+            y_offset = 0
+            for img in cropped_images:
+                collage_image[y_offset:y_offset+img.shape[0], :img.shape[1]] = img
+                y_offset += img.shape[0]
+
+            # Generate the combined filename from the file stem
+            collage_filename = f"{file_stem}_collage.jpg"
+            
+            # Save the collage image
+            collage_destination = os.path.join(Dirs.save_per_annotation_class, 'label', collage_filename)
+            validate_dir(os.path.dirname(collage_destination))
+            cv2.imwrite(collage_destination, collage_image)
+            logger.info(f"Saved collage image: {collage_destination}")
+
+            # Save each individual image separately
+            for filename in group:
+                original_image_name = os.path.basename(filename)
+                save_destination = os.path.join(Dirs.save_original, original_image_name)
+                validate_dir(os.path.dirname(save_destination))
+                cv2.imwrite(save_destination, cv2.imread(os.path.join(Dirs.save_original, filename)))
+                logger.info(f"Saved original image: {save_destination}")
 
 def crop_component_from_yolo_coords(anno_type, Dirs, analysis, all_detections, full_image, filename, save_per_image, save_per_class, save_list):
     height = analysis['height']

vouchervision/vouchervision_main.py

@@ -8,7 +8,7 @@ from time import perf_counter
 # sys.path.append(parentdir)
 # sys.path.append(currentdir)
 from vouchervision.component_detector.component_detector import detect_plant_components, detect_archival_components
-from vouchervision.general_utils import save_token_info_as_csv, print_main_start, check_for_subdirs_VV, load_config_file, load_config_file_testing, report_config, save_config_file, crop_detections_from_images_VV
+from vouchervision.general_utils import create_specimen_collage, save_token_info_as_csv, print_main_start, check_for_subdirs_VV, load_config_file, load_config_file_testing, report_config, save_config_file, crop_detections_from_images_VV
 from vouchervision.directory_structure_VV import Dir_Structure
 from vouchervision.data_project import Project_Info
 from vouchervision.LM2_logger import start_logging
@@ -62,6 +62,10 @@ def voucher_vision(cfg_file_path, dir_home, path_custom_prompts, cfg_test, progr
     # Save cropped detections
     crop_detections_from_images_VV(cfg, logger, dir_home, Project, Dirs)
 
+    
+    create_specimen_collage(cfg, logger, dir_home, Project, Dirs)
+
+
     # Process labels
     Voucher_Vision = VoucherVision(cfg, logger, dir_home, path_custom_prompts, Project, Dirs, is_hf)
     n_images = len(Voucher_Vision.img_paths)