shivahanifi commented 10 months ago

My idea is to categorize the prompts into distinct groups based on the information they include. Each category will include the information as follows:

Category 1: "Pass the " + Object_Pose + " " + Object_Color + " " + Object_Name + " " + Object_Placement + "."
Category 2: "Pass the " + Object_Pose + " " + Object_Name + " " + Object_Placement + "."
Category 3: "Pass the " + Object_Color + " " + Object_Name + "."
Category 4: "Pass the " + Object_Name + "."

Sentence formation:

TBD: Another thing to consider is the verb and the form of the sentence. As observed earlier 'Pass me' makes the code confuse, however, only 'pass' can also be confusing since it implies to move the object to the direction specified.

shivahanifi commented 10 months ago

The caption categories are designed and added using the following code:

parser = argparse.ArgumentParser(description='Caption format selection')
parser.add_argument('--caption', type=int, choices=[1, 2, 3, 4], default=1, help='Specify a value (1, 2, 3, 4) to determine the caption type. 1 being the most detailed and 4 being least detailed.')
args=parser.parse_args()

file_path = "/home/suka/code/Data/annotated_MDETR_test_data"
folders = sorted([f for f in os.listdir(file_path) if os.path.isdir(os.path.join(file_path, f))])
#folders = folders[2:]
for folder in folders:
    folder_path = os.path.join(file_path, folder)
    rgb_folders_path = os.path.join(folder_path, 'rgb_img')
    normMap_folders_path = os.path.join(folder_path, 'normMap')
    rgb_folders = sorted([f for f in os.listdir(rgb_folders_path) if os.path.isdir(os.path.join(rgb_folders_path, f))])
    normMap_folders = sorted([f for f in os.listdir(normMap_folders_path) if os.path.isdir(os.path.join(normMap_folders_path, f))])
    for i in range(min(len(rgb_folders),len(normMap_folders))-1):
        # input images
        images_path = os.path.join(rgb_folders_path,rgb_folders[i])
        images = sorted([f for f in os.listdir(images_path) if '.xml' not in f])
        # input caption
        annotation_path = os.path.join(images_path, 'annotation.xml')
        tree = ET.parse(annotation_path)
        root = tree.getroot()
        for obj in root.findall('object'):
            if obj.find('name').text != 'head':
                obj_info = {
                    'name': obj.find('name').text,
                    'color': obj.find('color').text,
                    'pose': obj.find('pose').text,
                    'placement': obj.find('placement').text,
                    'bndbox': {
                        'xmin': int(obj.find('bndbox/xmin').text),
                        'ymin': int(obj.find('bndbox/ymin').text),
                        'xmax': int(obj.find('bndbox/xmax').text),
                        'ymax': int(obj.find('bndbox/ymax').text)
                    }
                }
        # input heatmaps        
        normMaps_path = os.path.join(normMap_folders_path, normMap_folders[i])
        normMaps = sorted([f for f in os.listdir(normMaps_path)])
        for j in range(min(len(images),len(normMaps))-1):
            im_path = os.path.join(images_path, images[j])
            im = Image.open(im_path)
            im.show()
            if args.caption == 1:
                caption = "Pass the " + obj_info['pose'] + " " + obj_info['color'] + " " + obj_info['name'] + " " + obj_info['placement'] + "."
            elif args.caption == 2:
                caption = "Pass the " + obj_info['pose'] + " " + obj_info['name'] + " " + obj_info['placement'] + "."
            elif args.caption == 3:
                caption = "Pass the " + obj_info['color'] + " " + obj_info['name'] + "."
            elif args.caption == 4:
                caption = "Pass the " + obj_info['name'] + "."

shivahanifi commented 10 months ago

The final version of the caption has distinct prompts based on the category and level of detail included.

caption categories `-cc`

A. The B. This is a C. Look at the D. Point at the E. Pass the

caption details `-cd`

pose + color + name + placement
pose + name+ placement
color + name
name

Detail\Category	A	B	C	D	E
1	"The " + Object_Pose + " " + Object_Color + " " + Object_Name + " " + Object_Placement + "."	"This is a " + Object_Pose + " " + Object_Color + " " + Object_Name + " " + Object_Placement + "."	"Look at the " + Object_Pose + " " + Object_Color + " " + Object_Name + " " + Object_Placement + "."	"Point at the " + Object_Pose + " " + Object_Color + " " + Object_Name + " " + Object_Placement + "."	"Pass the " + Object_Pose + " " + Object_Color + " " + Object_Name + " " + Object_Placement + "."
2	"The " + Object_Pose + " " + Object_Name + " " + Object_Placement + "."	"This is a " + Object_Pose + " " + Object_Name + " " + Object_Placement + "."	"Look at the " + Object_Pose + " " + Object_Name + " " + Object_Placement + "."	"Point at the " + Object_Pose + " " + Object_Name + " " + Object_Placement + "."	"Pass the " + Object_Pose + " " + Object_Name + " " + Object_Placement + "."
3	"The " + Object_Color + " " + Object_Name + "."	"This is a " + Object_Color + " " + Object_Name + "."	"Look at the " + Object_Color + " " + Object_Name + "."	"Point at the " + Object_Color + " " + Object_Name + "."	"Pass the " + Object_Color + " " + Object_Name + "."
4	"The " + Object_Name + "."	"This is a " + Object_Name + "."	"Look at the " + Object_Name + "."	"Point at the " + Object_Name + "."	"Pass the " + Object_Name + "."

shivahanifi commented 10 months ago

Initial code implementation:

parser = argparse.ArgumentParser(description='Caption format selection')
parser.add_argument('-cc', '--caption_category', type=str, choices=['A', 'B', 'C', 'D', 'E'], default='A', help='Specify a value (A, B, C, D, E) to determine the caption category. A:The, B:This is a, C:Look at the, D:Point at the, E:Pass the')
parser.add_argument('-cd', '--caption_details', type=int, choices=[1, 2, 3, 4], default=1, help='Specify a detail level as (1, 2, 3, 4) to determine the caption details. 1:pose+color+name+placement, 2:pose+name+placement, 3:color+name, 4:name')
args=parser.parse_args()

# Caption creation
            caption_category = args.caption_category
            caption_details = args.caption_details
            if caption_category == 'A':
                if caption_details == 1:
                    caption = "The " + obj_info['pose'] + " " + obj_info['color'] + " " + obj_info['name'] + " " + obj_info['placement'] + "."
                elif caption_details == 2:
                    caption = "The " + obj_info['pose'] + " " + obj_info['name'] + " " + obj_info['placement'] + "."
                elif caption_details == 3:
                    caption = "The " + obj_info['color'] + " " + obj_info['name'] + "."
                elif caption_details == 4:
                    caption = "The " + obj_info['name'] + "."        
            elif caption_category == 'B':
                if caption_details == 1:
                    caption = "This is a " + obj_info['pose'] + " " + obj_info['color'] + " " + obj_info['name'] + " " + obj_info['placement'] + "."
                elif caption_details == 2:
                    caption = "This is a " + obj_info['pose'] + " " + obj_info['name'] + " " + obj_info['placement'] + "."
                elif caption_details == 3:
                    caption = "This is a " + obj_info['color'] + " " + obj_info['name'] + "."
                elif caption_details == 4:
                    caption = "This is a " + obj_info['name'] + "."        
            elif caption_category == 'C':
                if caption_details == 1:
                    caption = "Look at the " + obj_info['pose'] + " " + obj_info['color'] + " " + obj_info['name'] + " " + obj_info['placement'] + "."
                elif caption_details == 2:
                    caption = "Look at the " + obj_info['pose'] + " " + obj_info['name'] + " " + obj_info['placement'] + "."
                elif caption_details == 3:
                    caption = "Look at the " + obj_info['color'] + " " + obj_info['name'] + "."
                elif caption_details == 4:
                    caption = "Look at the " + obj_info['name'] + "." 
            elif caption_category == 'D':
                if caption_details == 1:
                    caption = "Point at the " + obj_info['pose'] + " " + obj_info['color'] + " " + obj_info['name'] + " " + obj_info['placement'] + "."
                elif caption_details == 2:
                    caption = "Point at the " + obj_info['pose'] + " " + obj_info['name'] + " " + obj_info['placement'] + "."
                elif caption_details == 3:
                    caption = "Point at the " + obj_info['color'] + " " + obj_info['name'] + "."
                elif caption_details == 4:
                    caption = "Point at the " + obj_info['name'] + "."    
            elif caption_category == 'E':
                if caption_details == 1:
                    caption = "Pass the " + obj_info['pose'] + " " + obj_info['color'] + " " + obj_info['name'] + " " + obj_info['placement'] + "."
                elif caption_details == 2:
                    caption = "Pass the " + obj_info['pose'] + " " + obj_info['name'] + " " + obj_info['placement'] + "."
                elif caption_details == 3:
                    caption = "Pass the " + obj_info['color'] + " " + obj_info['name'] + "."
                elif caption_details == 4:
                    caption = "Pass the " + obj_info['name'] + "."   
            print('Caption: ', caption)

shivahanifi commented 10 months ago

Optimized code implementation

# Define caption templates
            caption_templates = {
                'A': {
                    1: "The {pose} {color} {name} {placement}.",
                    2: "The {pose} {name} {placement}.",
                    3: "The {color} {name}.",
                    4: "The {name}.",
                },
                'B': {
                    1: "This is a {pose} {color} {name} {placement}.",
                    2: "This is a {pose} {name} {placement}.",
                    3: "This is a {color} {name}.",
                    4: "This is a {name}.",
                },
                'C': {
                    1: "Look at the {pose} {color} {name} {placement}.",
                    2: "Look at the {pose} {name} {placement}.",
                    3: "Look at the {color} {name}.",
                    4: "Look at the {name}.",
                },
                'D': {
                    1: "Point at the {pose} {color} {name} {placement}.",
                    2: "Point at the {pose} {name} {placement}.",
                    3: "Point at the {color} {name}.",
                    4: "Point at the {name}.",
                },
                'E': {
                    1: "Pass the {pose} {color} {name} {placement}.",
                    2: "Pass the {pose} {name} {placement}.",
                    3: "Pass the {color} {name}.",
                    4: "Pass the {name}.",
                }
            }

            # Construct caption
            caption_category = args.caption_category
            caption_details = args.caption_details
            caption = caption_templates[caption_category][caption_details].format(**obj_info)
            print('caption: ', caption)

shivahanifi / GazeMDETR

Generate prompts from annotation XML files #13

Sentence formation:

caption categories `-cc`

caption details `-cd`

Initial code implementation:

Optimized code implementation

shivahanifi / GazeMDETR

Generate prompts from annotation XML files #13

Sentence formation:

caption categories -cc

caption details -cd

Initial code implementation:

Optimized code implementation

caption categories `-cc`

caption details `-cd`