AutoTrain / app.py
ariankhalfani's picture
Update app.py
71f851a verified
import re
def generate_script_v8(dataset_code, task, model_size, epochs, batch_size):
# Extract the necessary information from the dataset code
api_key_match = re.search(r'api_key="(.*?)"', dataset_code)
workspace_match = re.search(r'workspace\("([^"]+)"\)', dataset_code)
project_name_match = re.search(r'project\("([^"]+)"\)', dataset_code)
version_number_match = re.search(r'version\((\d+)\)', dataset_code)
if not (api_key_match and workspace_match and project_name_match and version_number_match):
return "Error: Could not extract necessary information from the dataset code."
api_key = api_key_match.group(1)
workspace = workspace_match.group(1)
project_name = project_name_match.group(1)
version_number = int(version_number_match.group(1))
# Determine the model type based on the selected task
model_type = "seg" if task == "Segmentation" else "cls"
# Generate the script
script = f"""
import yaml
from ultralytics import YOLO
from roboflow import Roboflow
import logging
import re
import threading
import time
from io import StringIO
# Set up logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)
def auto_train():
log_stream = StringIO()
log_handler = logging.StreamHandler(log_stream)
log_handler.setLevel(logging.INFO)
logger.addHandler(log_handler)
try:
api_key = "{api_key}"
workspace = "{workspace}"
project_name = "{project_name}"
version_number = {version_number}
# Load the Roboflow dataset
rf = Roboflow(api_key=api_key)
project = rf.workspace(workspace).project(project_name)
version = project.version(version_number)
dataset = version.download("yolov8")
# Modify the data structure
yaml_file_path = f'{{dataset.location}}/data.yaml'
with open(yaml_file_path, 'r') as file:
data = yaml.safe_load(file)
data['val'] = '../valid/images'
data['test'] = '../test/images'
data['train'] = '../train/images'
with open(yaml_file_path, 'w') as file:
yaml.safe_dump(data, file)
# Determine the model name based on the selected size and task
model_name = f"yolov8{model_size}-{model_type}.pt"
# Load and train the model
model = YOLO(model_name)
model.info()
# Function to read logs in real-time and update the Streamlit textbox
def update_logs():
while getattr(threading.currentThread(), "do_run", True):
time.sleep(1)
log_stream.seek(0)
print(log_stream.read())
# Start a thread to update logs in real-time
log_thread = threading.Thread(target=update_logs)
log_thread.start()
results = model.train(data=yaml_file_path, epochs={epochs}, imgsz=640, batch={batch_size})
# Stop the log update thread
logger.removeHandler(log_handler)
log_thread.do_run = False
log_thread.join()
# Return the result path and logs
log_stream.seek(0)
log_output = log_stream.read()
print("Results Directory:", results.results_dir)
print("Final Training Logs:", log_output)
except Exception as e:
logger.error(f"An error occurred: {{e}}")
log_stream.seek(0)
log_output = log_stream.read()
print(f"Error: {{e}}")
print(log_output)
finally:
logger.removeHandler(log_handler)
if __name__ == "__main__":
auto_train()
"""
return script
def generate_script_v9(dataset_code, task, model_size, epochs, batch_size):
# Extract the necessary information from the dataset code
api_key_match = re.search(r'api_key="(.*?)"', dataset_code)
workspace_match = re.search(r'workspace\("([^"]+)"\)', dataset_code)
project_name_match = re.search(r'project\("([^"]+)"\)', dataset_code)
version_number_match = re.search(r'version\((\d+)\)', dataset_code)
if not (api_key_match and workspace_match and project_name_match and version_number_match):
return "Error: Could not extract necessary information from the dataset code."
api_key = api_key_match.group(1)
workspace = workspace_match.group(1)
project_name = project_name_match.group(1)
version_number = int(version_number_match.group(1))
# Determine the model name based on the selected size and task
if task == "Segmentation":
model_name = f"gelan-c-seg.pt" if model_size == "c" else f"yolov9-{model_size}-seg.pt"
else:
model_name = f"yolov9-{model_size}.pt"
# Generate the script
script = f"""
!pip install roboflow
from roboflow import Roboflow
rf = Roboflow(api_key="{api_key}")
project = rf.workspace("{workspace}").project("{project_name}")
version = project.version({version_number})
dataset = version.download("yolov9")
!python train.py \\
--batch {batch_size} --epochs {epochs} --img 640 --device 0 --min-items 0 --close-mosaic 15 \\
--data {{dataset.location}}/data.yaml \\
--weights {{HOME}}/weights/{model_name} \\
--cfg models/detect/{model_name.split('.')[0]}.yaml \\
--hyp hyp.scratch-high.yaml
"""
return script
import streamlit as st
st.title("Auto Train Script Generator")
st.write("Generate a YOLO training script using a Roboflow dataset")
tab1, tab2 = st.tabs(["YOLOv8", "YOLOv9"])
with tab1:
st.subheader("YOLOv8 Script Generator")
dataset_code = st.text_input("Roboflow Dataset Code", key="dataset_code_v8", placeholder="Paste your Roboflow dataset code here")
task = st.selectbox("Task", ["Object Detection", "Segmentation"], index=0, key="task_v8")
model_size = st.selectbox("Model Size", ["n", "s", "m", "l", "x"], index=0, key="model_size_v8")
epochs = st.selectbox("Epochs", [50, 100, 200, 300, 400, 500], index=3, key="epochs_v8")
batch_size = st.selectbox("Batch Size", [1, 2, 4, 8, 16, 32], index=0, key="batch_size_v8")
if st.button("Generate YOLOv8 Script"):
script = generate_script_v8(dataset_code, task, model_size, epochs, batch_size)
st.code(script, language="python")
with tab2:
st.subheader("YOLOv9 Script Generator")
dataset_code = st.text_input("Roboflow Dataset Code", key="dataset_code_v9", placeholder="Paste your Roboflow dataset code here")
task = st.selectbox("Task", ["Object Detection", "Segmentation"], index=0, key="task_v9")
model_size = st.selectbox("Model Size", ["t", "s", "m", "c", "e"], index=0, key="model_size_v9")
epochs = st.selectbox("Epochs", [50, 100, 200, 300, 400, 500], index=3, key="epochs_v9")
batch_size = st.selectbox("Batch Size", [1, 2, 4, 8, 16, 32], index=0, key="batch_size_v9")
if st.button("Generate YOLOv9 Script"):
script = generate_script_v9(dataset_code, task, model_size, epochs, batch_size)
st.code(script, language="python")