hydata / create_metadata_360_icl_v3.py
SCMayS's picture
Upload create_metadata_360_icl_v3.py with huggingface_hub
002ea66 verified
import os
import re
import glob
import json
import random
import argparse
from tqdm import tqdm
from pathlib import Path
from collections import defaultdict
def parse_ground_truth(name):
"""Extract ground truth rotation axis and angle from filename or folder name"""
# Remove file extension if present
basename = name.split(".")[0] if "." in name else name
parts = basename.split("_")
if len(parts) >= 4: # figXXXX_XXX_axis_angle
rotation_axis = parts[-2] # Second to last element is axis
rotation_angle = int(parts[-1]) # Last element is angle
# Convert negative angles to 0-360 range
if rotation_angle < 0:
rotation_angle += 360
return rotation_axis, rotation_angle
print(f"Warning: Could not parse name: {basename}")
return None, None
def load_examples(example_dir, generation_mode):
"""Load example images from the example directory"""
if generation_mode == "combined":
# Load all single PNG files from the example directory
files = glob.glob(os.path.join(example_dir, "*.png"))
print(f"Found {len(files)} combined example images in {example_dir}")
return files
else: # separate mode
# Find all folders in the example directory
folders = [f for f in glob.glob(os.path.join(example_dir, "*")) if os.path.isdir(f)]
# Filter folders that contain both _ini.png and _rot.png files
valid_folders = []
for folder in folders:
folder_name = os.path.basename(folder)
ini_file = os.path.join(folder, f"{folder_name}_ini.png")
rot_file = os.path.join(folder, f"{folder_name}_rot.png")
if os.path.exists(ini_file) and os.path.exists(rot_file):
valid_folders.append(folder)
print(f"Found {len(valid_folders)} example folder pairs in {example_dir}")
return valid_folders
def organize_examples(examples, generation_mode):
"""Organize examples by rotation axis and angle"""
organized = defaultdict(list)
for example in examples:
basename = os.path.basename(example)
if generation_mode == "combined":
basename = basename.split(".")[0]
axis, angle = parse_ground_truth(basename)
if axis is None or angle is None:
continue
key = (axis, angle)
organized[key].append(example)
# Print statistics
print("\nDistribution of examples by axis-angle:")
for key, examples_list in organized.items():
print(f" {key[0]}-axis, {key[1]} degrees: {len(examples_list)} examples")
return dict(organized)
def select_examples(organized_examples, test_axis, possible_angles, max_examples=1):
"""Select a single random example for the test case"""
# Collect all examples for this axis regardless of angle
all_examples_for_axis = []
for (axis, angle), example_list in organized_examples.items():
if axis == test_axis:
for example in example_list:
all_examples_for_axis.append((example, angle))
# If we have any examples for this axis, select one randomly
examples = []
if all_examples_for_axis:
selected_example = random.choice(all_examples_for_axis)
examples.append(selected_example)
else:
print(f"Warning: No examples found for rotation around {test_axis}-axis")
return examples
def construct_icl_prompt(axis, possible_angles, icl_examples, difficulty="easy", generation_mode="combined"):
"""Create prompt with in-context learning examples for the VLM"""
# Generate list of all possible rotation angles based on angle increment
'''possible_angles = []
current_angle = 0 + angle_increment
while current_angle < 360:
possible_angles.append(current_angle)
current_angle += angle_increment'''
# Common instructions for both modes
coordinate_system = (
f"The 3D Cartesian coordinate system is defined as follows: "
f"\n- x-axis: points horizontally from left to right (positive direction is right)"
f"\n- y-axis: points vertically from bottom to top (positive direction is up)"
f"\n- z-axis: points from inside the image toward the viewer (positive direction is out of the screen)"
f"\n- The origin (0,0,0) is located at the geometric center of the 3D object mesh"
f"\n\nWhen discussing rotations around an axis, imagine looking along the positive direction of that axis (as if looking from the origin toward the positive end)."
)
angle_constraints = (
f"The rotation angle is one of these specific values: {possible_angles} degrees. "
f"A positive angle means rotation in the CLOCKWISE direction when looking along the positive direction of the axis. "
)
# Create example section of the prompt (singular!)
examples_text = "\n### EXAMPLE OF ROTATION ###\n"
for idx, (_, angle) in enumerate(icl_examples):
if generation_mode == "combined":
img_num = idx + 1
examples_text += f"\nExample: Image {img_num} shows a 3D object with its left half showing the initial view and right half showing a {angle} degree rotation around the {axis}-axis.\n"
else: # separate mode
img_num_start = idx * 2 + 1
img_num_end = idx * 2 + 2
examples_text += f"\nExample: Image {img_num_start} shows the initial view and Image {img_num_end} shows the object after a {angle} degree rotation around the {axis}-axis.\n"
# Different instructions based on difficulty
if difficulty == "easy":
# For easy mode - axis is provided
thinking_instructions = (
f"IMPORTANT: Please follow this systematic approach to determine the rotation angle:"
f"\n\n1. First, analyze the object's features in both views to understand its structure."
f"\n\n2. For the {axis}-axis rotation, you must evaluate ONLY these possible rotation angles: {possible_angles}"
f"\n - For each angle in the list, describe what the object would look like after rotating around the {axis}-axis by that amount"
f"\n - Compare these descriptions with the actual second view"
f"\n - DO NOT make a decision until you have evaluated all possible angles in the list"
f"\n\n3. After evaluating all angles, choose the one that best matches the observed changes"
f"\n\n4. Verify your answer by mentally applying the rotation to confirm it matches the second view"
)
response_format = (
f"Place your detailed reasoning process in <think></think> tags. Your reasoning should include:"
f"\n- Systematic evaluation of possible rotation angles from the provided list"
f"\n- Specific visual features you used to determine your answer"
f"\n\nThen provide your final answer in <rotation_angle></rotation_angle> tags (use only a number from the list for angle)."
f"\ni.e., <think> your reasoning process here </think><rotation_angle> your predicted degrees here </rotation_angle>"
)
task_description = (
f"Your task is to determine the angle of rotation around the {axis}-axis in degrees."
)
else: # hard mode - axis is not provided
thinking_instructions = (
f"IMPORTANT: Please follow this systematic approach to determine the rotation:"
f"\n\n1. First, analyze the object's features in both views to understand its structure."
f"\n\n2. Consider what would happen if rotation occurred around each of the three axes (x, y, and z):"
f"\n - For x-axis rotation: What specific features would change and how?"
f"\n - For y-axis rotation: What specific features would change and how?"
f"\n - For z-axis rotation: What specific features would change and how?"
f"\n - Based on the observed changes, explain which axis makes the most sense and why."
f"\n\n3. Once you've determined the most likely axis, evaluate ALL of these possible rotation angles: {possible_angles}"
f"\n - For each angle in the list, describe what the object would look like after rotating around your chosen axis by that amount"
f"\n - Compare these descriptions with the actual second view"
f"\n - DO NOT make a decision until you have evaluated all angles in the list"
f"\n\n4. After evaluating all angles, choose the one that best matches the observed changes"
)
response_format = (
f"Place your detailed reasoning process in <think></think> tags. Your reasoning should include:"
f"\n- Analysis of how rotation around each axis would affect the object"
f"\n- Systematic evaluation of possible rotation angles from the provided list"
f"\n- Specific visual features you used to determine your answer"
f"\n\nThen provide your final answer in <rotation_axis></rotation_axis> and <rotation_angle></rotation_angle> tags respectively (use only x, y, or z for axis and only a number from the list for angle)."
f"\ni.e., <think> your reasoning process here </think><rotation_axis> your predicted axis here </rotation_axis><rotation_angle> your predicted degrees here </rotation_angle>"
)
task_description = (
f"Your task is to determine which axis the object was rotated around and by what angle."
)
# Generate the prompt based on generation mode
if generation_mode == "combined":
test_img_num = len(icl_examples) + 1
prompt = (
f"IMPORTANT: I'm showing you multiple images of 3D objects. "
f"The test case (final image) contains TWO separate 3D renderings side-by-side. "
f"\n\nThe LEFT HALF shows a 3D object in its initial orientation. "
f"The RIGHT HALF shows the SAME 3D object after being rotated. "
f"\n\nYour task is to determine the angle of rotation around the {axis}-axis in degrees."
f"\n\n{coordinate_system}"
f"\n\n{angle_constraints}"
f"\n\n{examples_text}"
f"\n\n### YOUR TASK ###"
f"\nNow, for Image {test_img_num}, determine the angle of rotation around the {axis}-axis."
f"\nBased on the examples provided, analyze Image {test_img_num} carefully."
f"\n\n{thinking_instructions}"
f"\n\n{response_format}"
)
else: # separate mode
test_img_start = len(icl_examples) * 2 + 1
test_img_end = len(icl_examples) * 2 + 2
prompt = (
f"I'm showing you multiple images of 3D objects. "
f"For each example or test case, two images represent the same object before and after rotation."
f"\n\nYour task is to determine the angle of rotation around the {axis}-axis in degrees."
f"\n\n{coordinate_system}"
f"\n\n{angle_constraints}"
f"\n\n{examples_text}"
f"\n\n### YOUR TASK ###"
f"\nNow, determine the angle of rotation around the {axis}-axis from Image {test_img_start} to Image {test_img_end}."
f"\nBased on the examples provided, analyze the rotation carefully."
f"\n\n{thinking_instructions}"
f"\n\n{response_format}"
)
return prompt
def create_metadata_jsonl_combined_icl(input_dir, example_dir, output_file, possible_angles=[45, 315], difficulty="easy", max_examples=3):
"""Create metadata JSONL file for all images in input_dir with in-context learning examples (combined mode)"""
# Get all PNG files in the input directory
png_files = glob.glob(os.path.join(input_dir, "*.png"))
# Sort files to ensure consistent order
png_files = sorted(png_files)
if not png_files:
print(f"No PNG files found in {input_dir}")
return
print(f"Found {len(png_files)} PNG files in {input_dir}")
# Load example images
examples = load_examples(example_dir, "combined")
# Organize examples by axis and angle
organized_examples = organize_examples(examples, "combined")
# Create output directory if it doesn't exist
output_dir = os.path.dirname(output_file)
os.makedirs(output_dir, exist_ok=True)
# Get the last folder name from the input directory
last_folder = os.path.basename(os.path.normpath(input_dir))
# Define the target base directory
target_base_dir = f"/lustre/fsw/portfolios/av/users/shiyil/yunfei/MM-EUREKA/data/{last_folder}"
# Example directory - assume it's the "examples" subdirectory of input_dir
target_example_dir = f"{target_base_dir}/examples"
# Generate list of all possible rotation angles
'''possible_angles = []
current_angle = 0 + angle_increment
while current_angle < 360:
possible_angles.append(current_angle)
current_angle += angle_increment'''
# Process each file and create metadata entries
entries = []
for png_file in tqdm(png_files, desc="Creating metadata for combined mode with ICL"):
# Parse ground truth from filename
axis, angle = parse_ground_truth(os.path.basename(png_file))
if axis is None or angle is None:
print(f"Skipping {png_file} - could not parse ground truth")
continue
# Get the basename
basename = os.path.basename(png_file)
# Create the new image path for the target system
target_image_path = f"{target_base_dir}/{basename}"
# Select in-context examples for this test case
icl_examples = select_examples(organized_examples, axis, possible_angles, max_examples)
# Skip if no examples found
if not icl_examples:
print(f"Skipping {png_file} - no suitable examples found")
continue
# Construct prompt with in-context examples
prompt = construct_icl_prompt(axis, possible_angles, icl_examples, difficulty, generation_mode="combined")
# Create answer format based on difficulty WITH image path
if difficulty == "easy":
# For easy mode, only include angle in the answer (axis is provided in the prompt)
answer = f"<angle>{angle}</angle><image_path>{target_image_path}</image_path>"
else:
# For hard mode, include both axis and angle
answer = f"<axis>{axis}</axis><angle>{angle}</angle><image_path>{target_image_path}</image_path>"
# Create content array with example images and test image
content = []
# Add example images
for example_path, _ in icl_examples:
example_basename = os.path.basename(example_path)
target_example_path = f"{target_example_dir}/{example_basename}"
content.append({"type": "image", "image": target_example_path})
# Add test image
content.append({"type": "image", "image": target_image_path})
# Add prompt text
content.append({"type": "text", "text": prompt})
# Create entry with the content array
entry = {
"message": json.dumps([{
"role": "user",
"content": content
}]),
"answer": answer
}
entries.append(entry)
# Write entries to JSONL file
with open(output_file, 'w') as f:
for entry in entries:
f.write(json.dumps(entry) + '\n')
print(f"\nSummary for combined mode with ICL:")
print(f" Found {len(png_files)} PNG files")
print(f" Created metadata for {len(entries)} entries")
print(f" Output file: {output_file}")
print(f" Image paths are set to: {target_base_dir}/[filename].png")
print(f" Example paths are set to: {target_example_dir}/[filename].png")
def create_metadata_jsonl_separate_icl(input_dir, example_dir, output_file, possible_angles=[45, 315], difficulty="easy", max_examples=3):
"""Create metadata JSONL file for folders in input_dir with in-context learning examples (separate mode)"""
# Get all directories in the input directory (excluding the examples directory)
folders = [f for f in glob.glob(os.path.join(input_dir, "*"))
if os.path.isdir(f) and os.path.basename(f) != "examples"]
# Sort folders to ensure consistent order
folders = sorted(folders)
if not folders:
print(f"No folders found in {input_dir}")
return
print(f"Found {len(folders)} folders in {input_dir}")
# Load example folders
examples = load_examples(example_dir, "separate")
# Organize examples by axis and angle
organized_examples = organize_examples(examples, "separate")
# Create output directory if it doesn't exist
output_dir = os.path.dirname(output_file)
os.makedirs(output_dir, exist_ok=True)
# Get the last folder name from the input directory
last_folder = os.path.basename(os.path.normpath(input_dir))
# Define the target base directory
target_base_dir = f"/lustre/fsw/portfolios/av/users/shiyil/yunfei/MM-EUREKA/data/{last_folder}"
# Example directory is always a subdirectory called 'examples'
target_example_dir = f"{target_base_dir}/examples"
# Generate list of all possible rotation angles
'''possible_angles = []
current_angle = 0 + angle_increment
while current_angle < 360:
possible_angles.append(current_angle)
current_angle += angle_increment'''
# Process each folder and create metadata entries
entries = []
valid_folders = 0
for folder in tqdm(folders, desc="Creating metadata for separate mode with ICL"):
folder_name = os.path.basename(folder)
# Parse ground truth from folder name
axis, angle = parse_ground_truth(folder_name)
if axis is None or angle is None:
print(f"Skipping {folder} - could not parse ground truth")
continue
# Check for the two required images in the folder
ini_path = os.path.join(folder, f"{folder_name}_ini.png")
rot_path = os.path.join(folder, f"{folder_name}_rot.png")
if not os.path.exists(ini_path):
print(f"Skipping {folder} - missing initial view image")
continue
if not os.path.exists(rot_path):
print(f"Skipping {folder} - missing rotated view image")
continue
# Create target paths for remote system
target_folder_path = f"{target_base_dir}/{folder_name}"
target_ini_path = f"{target_folder_path}/{folder_name}_ini.png"
target_rot_path = f"{target_folder_path}/{folder_name}_rot.png"
# Select in-context examples for this test case
icl_examples = select_examples(organized_examples, axis, possible_angles, max_examples)
# Skip if no examples found
if not icl_examples:
print(f"Skipping {folder} - no suitable examples found")
continue
# Construct prompt with in-context examples
prompt = construct_icl_prompt(axis, possible_angles, icl_examples, difficulty, generation_mode="separate")
# Create answer format based on difficulty WITH folder path
if difficulty == "easy":
# For easy mode, only include angle in the answer (axis is provided in the prompt)
answer = f"<angle>{angle}</angle><image_path>{target_folder_path}</image_path>"
else:
# For hard mode, include both axis and angle
answer = f"<axis>{axis}</axis><angle>{angle}</angle><image_path>{target_folder_path}</image_path>"
# Create content array with example images and test images
content = []
# Add example image pairs
for example_folder, _ in icl_examples:
example_name = os.path.basename(example_folder)
target_example_ini = f"{target_example_dir}/{example_name}/{example_name}_ini.png"
target_example_rot = f"{target_example_dir}/{example_name}/{example_name}_rot.png"
content.append({"type": "image", "image": target_example_ini})
content.append({"type": "image", "image": target_example_rot})
# Add test image pair
content.append({"type": "image", "image": target_ini_path})
content.append({"type": "image", "image": target_rot_path})
# Add prompt text
content.append({"type": "text", "text": prompt})
# Create entry with both image paths
entry = {
"message": json.dumps([{
"role": "user",
"content": content
}]),
"answer": answer
}
entries.append(entry)
valid_folders += 1
# Write entries to JSONL file
with open(output_file, 'w') as f:
for entry in entries:
f.write(json.dumps(entry) + '\n')
print(f"\nSummary for separate mode with ICL:")
print(f" Found {len(folders)} folders")
print(f" Created metadata for {valid_folders} valid folders")
print(f" Output file: {output_file}")
print(f" Image paths format: {target_base_dir}/[folder_name]/[folder_name]_[ini/rot].png")
print(f" Example paths format: {target_example_dir}/[folder_name]/[folder_name]_[ini/rot].png")
def main():
parser = argparse.ArgumentParser(description="Create metadata JSONL for rotation dataset with in-context learning")
parser.add_argument('--input-dir', type=str, required=True,
help="Directory containing rotation dataset images or folders")
parser.add_argument('--example-dir', type=str, required=True,
help="Directory containing example images or folders for in-context learning")
parser.add_argument('--output-file', type=str, default="rotation_metadata_icl.jsonl",
help="Output JSONL file path")
parser.add_argument('--possible-angles', type=int, nargs='+', default=[45, 315],
help="List of possible rotation angles in degrees (e.g., 45 315)")
parser.add_argument('--difficulty', type=str, choices=["easy", "hard"], default="easy",
help="Difficulty mode: easy (axis provided) or hard (axis not provided)")
parser.add_argument('--generation-mode', type=str, choices=["combined", "separate"], default="combined",
help="Mode for dataset generation (combined = one image with both views, separate = folder with two images)")
parser.add_argument('--max-examples', type=int, default=1,
help="Maximum number of examples to include for each test case (default: 1)")
parser.add_argument('--random-seed', type=int, default=None,
help="Random seed for example selection (None for true randomness)")
args = parser.parse_args()
# Set random seed for reproducibility (or use None for true randomness)
random.seed(args.random_seed)
print(f"Creating metadata JSONL for rotation dataset with in-context learning:")
print(f"Input directory: {args.input_dir}")
print(f"Example directory: {args.example_dir}")
print(f"Output file: {args.output_file}")
print(f"Possible angles: {args.possible_angles}")
print(f"Difficulty mode: {args.difficulty}")
print(f"Generation mode: {args.generation_mode}")
print(f"Max examples per test case: {args.max_examples}")
if args.random_seed is not None:
print(f"Using fixed random seed: {args.random_seed}")
else:
print("Using true randomness (different examples selection each run)")
# Fix the indentation error here:
if args.example_dir is None and os.path.exists(os.path.join(args.input_dir, "examples")):
args.example_dir = os.path.join(args.input_dir, "examples")
print(f"Using examples directory: {args.example_dir}")
if args.generation_mode == "combined":
create_metadata_jsonl_combined_icl(
input_dir=args.input_dir,
example_dir=args.example_dir,
output_file=args.output_file,
possible_angles=args.possible_angles, # Updated from angle_increment
difficulty=args.difficulty,
max_examples=args.max_examples
)
else: # separate mode
create_metadata_jsonl_separate_icl(
input_dir=args.input_dir,
example_dir=args.example_dir,
output_file=args.output_file,
possible_angles=args.possible_angles, # Updated from angle_increment
difficulty=args.difficulty,
max_examples=args.max_examples
)
if __name__ == "__main__":
main()