SushantGautam's picture
Add challenge evaluation support by patching submission file and processing predictions
1ba23a3
from gradio_client import Client, handle_file
from huggingface_hub import snapshot_download, login, whoami
import argparse
import os
import subprocess as sp
import time
from datetime import datetime, timezone
import shutil # Add this import
import json
from huggingface_hub import HfApi, grant_access
import re
HF_GATE_ACESSLIST = ["SushantGautam",
"stevenah", "vlbthambawita"]
MEDVQA_SUBMIT = True if os.environ.get(
'_MEDVQA_SUBMIT_FLAG_', 'FALSE') == 'TRUE' else False
parser = argparse.ArgumentParser(description='Run GI-1015 Task 1 (VQA)')
parser.add_argument('--repo_id', type=str, required=True,
help='Path to the HF submission repository')
args, _ = parser.parse_known_args()
os.environ["HF_HUB_ENABLE_HF_TRANSFER"] = "1"
submission_file = "submission_task1.py"
file_from_validation = "predictions_1.json"
min_library = ["datasets>=3.4.1", "transformers", "evaluate",
"rouge_score", 'tqdm', "gradio_client>=1.8.0"]
print("🌟 ImageCLEFmed-MEDVQA-GI-2025 🌟",
"https://github.com/simula/ImageCLEFmed-MEDVQA-GI-2025")
print("πŸ” Subtask 1: Algorithm Development for Question Interpretation and Response")
print(f"πŸ‘€ Analyzing submission repository: {args.repo_id} πŸ‘€")
try:
print(f"Logged in to HuggingFace as: {whoami()['name']}")
except Exception:
print("⚠️⚠️ Not logged in to HuggingFace! Please get your login token from https://huggingface.co/settings/tokens 🌐")
login()
client = Client("SimulaMet/medvqa")
print("πŸ’“ Communicating with the Submission Server: Ping!")
result = client.predict(
api_name="/refresh_page"
)
print(result)
hf_username = whoami()['name']
assert len(hf_username) > 0, "🚫 HuggingFace login failed for some reason"
current_timestamp = int(time.time())
snap_dir = snapshot_download(
repo_id=args.repo_id, allow_patterns=[submission_file, "requirements.txt"])
if not os.path.isfile(os.path.join(snap_dir, submission_file)):
raise FileNotFoundError(
f"Submission file '{submission_file}' not found in the repository!")
if os.path.isfile(os.path.join(snap_dir, file_from_validation)):
os.remove(os.path.join(snap_dir, file_from_validation))
print("πŸ“¦ Making sure of the minimum requirements to run the script πŸ“¦")
sp.run(["python", "-m", "pip", "install", "-q"] + min_library, check=True)
if os.path.isfile(os.path.join(snap_dir, "requirements.txt")):
print(
f"πŸ“¦ Installing requirements from the submission repo: {args.repo_id}/requirements.txt")
sp.run(["python", "-m", "pip", "install", "-q", "-r",
f"{snap_dir}/requirements.txt"], cwd=snap_dir, check=True)
if os.environ.get("_MEDVQA_CHALLENGE_EVALUATE_FLAG_", "FALSE") == "TRUE":
# Patch submission file for challenge evaluation
challenge_file = submission_file.replace(".py", "_challenge.py")
submission_path = os.path.join(snap_dir, submission_file)
challenge_path = os.path.join(snap_dir, challenge_file)
with open(submission_path, "r", encoding="utf-8") as f:
code = f.read()
# Use regex to match the line, ignoring whitespace
pattern = r'val_dataset\s*=\s*load_dataset\(\s*["\']SimulaMet/Kvasir-VQA-test["\']\s*,\s*split\s*=\s*["\']validation["\']\s*\)'
new_line = 'val_dataset = load_dataset("SimulaMet/Kvasir-VQA-private", split="test")'
if re.search(pattern, code):
code = re.sub(pattern, new_line, code)
with open(challenge_path, "w", encoding="utf-8") as f:
f.write(code)
submission_file = challenge_file
print(f"πŸ”„ Challenge file created at: {challenge_path}")
else:
print("⚠️ Challenge patch not applied: expected line not found in submission file.")
os.exit(
"Please check the submission file for compatibility with challenge evaluation.")
sp.run(["python", f"{snap_dir}/{submission_file}"],
cwd=snap_dir, check=True)
print(
f"πŸŽ‰ The submission script ran successfully, the intermediate files are at {snap_dir}")
if not MEDVQA_SUBMIT:
print("\n You can now run medvqa validate_and_submit .... command to submit the task.")
else:
print("πŸš€ Preparing for submission πŸš€")
file_path_to_upload = os.path.join(
snap_dir, f"{hf_username}-_-_-{current_timestamp}-_-_-task1.json")
shutil.copy(os.path.join(snap_dir, file_from_validation),
file_path_to_upload) # Use shutil.copy here
# add repo_id to the submission file
with open(file_path_to_upload, 'r', encoding='utf-8') as f:
data = json.load(f)
data['repo_id'] = args.repo_id
with open(file_path_to_upload, 'w', encoding='utf-8') as f:
json.dump(data, f, ensure_ascii=False)
api = HfApi()
api.update_repo_visibility(args.repo_id, private=False) # Make public
api.update_repo_settings(
args.repo_id, gated='manual') # Enable gated access
for user in HF_GATE_ACESSLIST:
try:
grant_access(args.repo_id, user) # Grant access
except Exception as e:
print(user, ":", e)
print(
f'''βœ… {args.repo_id} model is now made public, but gated, and is shared with organizers.
You should not make the model private or remove/update it until the competition results are announced.
Feel feel to re-submit the task if you change the model on the repository.
We will notify you if there are any issues with the submission.
''')
result = client.predict(
file=handle_file(file_path_to_upload),
api_name="/add_submission"
)
print({"User": hf_username, "Task": "task1",
"Submitted_time": str(datetime.fromtimestamp(int(current_timestamp), tz=timezone.utc)) + " UTC"
})
print(result)
print("Visit this URL to see the entry: πŸ‘‡")
Client("SimulaMet/medvqa")
if os.environ.get("_MEDVQA_CHALLENGE_EVALUATE_FLAG_", "FALSE") == "TRUE":
src_json = os.path.join(snap_dir, "predictions_1.json")
if os.path.isfile(src_json):
with open(src_json, "r", encoding="utf-8") as f:
data = json.load(f)
# Remove 'debug' key if present
data.pop("debug", None)
# Rename 'public_scores' to 'challenge_scores' if present
if "public_scores" in data:
data["challenge_scores"] = data.pop("public_scores")
# Get Team_Name from submission_info
team_name = data.get("submission_info", {}).get(
"Team_Name", "unknown_team")
team_name_safe = re.sub(r'[^a-zA-Z0-9_\-]', '_', team_name)
out_json = os.path.join(os.getcwd(), f"task1_{team_name_safe}.json")
with open(out_json, "w", encoding="utf-8") as f:
json.dump(data, f, ensure_ascii=False, indent=2)
print(f"βœ… Copied and processed predictions to: {out_json}")
else:
print("❌ predictions_1.json not found in snapshot directory!")
# === End: Post-processing predictions_1.json ===