Issue Summary: YOLOv8 Training Cache Problem

[aeromista2a]

Hello everyone,

I hope you're all doing well. My name is Louis, and I am a French men working on a custom pokercards detection project using YOLOv8. I am reaching out for assistance regarding an issue I am encountering with the cache files during the training process.
Problem:
I am facing issues with training a custom dataset using YOLOv8. Despite following the dataset formatting guidelines, the training process does not correctly utilize the cache files. The .cache files are created in the main directories (Images and Labels), but the model fails to use the cache files in the appropriate subdirectories (train, val, test). As a result, warnings about missing labels appear, and the training does not proceed as expected.

Steps Taken:

Verified that the dataset directory structure is correct.
Created a dataset.yaml file pointing to the correct paths for images and labels.
Ensured annotations exist in the respective directories.
Created cache files in the correct subdirectories.
Configured the training script to reference the correct paths.
Persistent Issue:
Despite these efforts, the YOLOv8 model still fails to recognize the cache files in the subdirectories, resulting in warnings and training issues.

Transferred 349/355 items from pretrained weights
Freezing layer 'model.22.dfl.conv.weight'
train: Scanning C:\Users\Louis\Desktop\Bot\dataset\Images\train.ca
WARNING ⚠️ No labels found in C:\Users\Louis\Desktop\Bot\dataset\Images\train.cache, training may not work correctly. See https://docs.ultralytics.com/datasets for dataset formatting guidance.
train: Caching images (0.1GB RAM): 100%|██████████| 74/74 [00:00<0
val: Scanning C:\Users\Louis\Desktop\Bot\dataset\Images\val.cache.
WARNING ⚠️ No labels found in C:\Users\Louis\Desktop\Bot\dataset\Images\val.cache, training may not work correctly. See https://docs.ultralytics.com/datasets for dataset formatting guidance.
val: Caching images (0.1GB RAM): 100%|██████████| 74/74 [00:00<00:
Plotting labels to runs\detect\poker_cards618\labels.jpg...
zero-size array to reduction operation maximum which has no identity
optimizer: 'optimizer=auto' found, ignoring 'lr0=0.01' and 'momentum=0.937' and determining best 'optimizer', 'lr0' and 'momentum' automatically...
optimizer: AdamW(lr=0.000476, momentum=0.9) with parameter groups 57 weight(decay=0.0), 64 weight(decay=0.0005), 63 bias(decay=0.0)
Image sizes 640 train, 640 val
Using 0 dataloader workers
Logging results to runs\detect\poker_cards618
Starting training for 50 epochs...

YOLOv8 Directory Structure
C:/Users/Louis/Desktop/Bot/
|-- dataset/
| |-- Images/
| | |-- train/
| | | |-- image_000001.jpg
| | | |-- image_000002.jpg
| | | |-- ...
| | | -- train.cache | | |-- val/ | | | |-- image_000001.jpg | | | |-- image_000002.jpg | | | |-- ... | | | -- val.cache
| | |-- test/
| | |-- image_000001.jpg
| | |-- image_000002.jpg
| | |-- ...
| | -- test.cache | |-- Labels/ | |-- train/ | | |-- image_000001.txt | | |-- image_000002.txt | | |-- ... | | -- train.cache
| |-- val/
| | |-- image_000001.txt
| | |-- image_000002.txt
| | |-- ...
| | -- val.cache | |-- test/ | |-- image_000001.txt | |-- image_000002.txt | |-- ... | -- test.cache
|
|-- dataset.yaml
|-- train_yolov8.py
|-- requirements.txt
|-- venv/
|-- ...

Example dataset.yaml
path: C:/Users/Louis/Desktop/Bot/dataset
train:
images: Images/train
labels: Labels/train
val:
images: Images/val
labels: Labels/val
test:
images: Images/test
labels: Labels/test

nc: 137

names:
0: a_coeur
1: 2_coeur
...
136: tapis j1 table de 8

Example Python Script (train_yolov8.py)

import os
from ultralytics import YOLO

Define paths

base_path = 'C:/Users/Louis/Desktop/Bot/dataset'
image_dirs = {
'train': os.path.join(base_path, "Images/train"),
'val': os.path.join(base_path, "Images/val"),
'test': os.path.join(base_path, "Images/test")
}

label_dirs = {
'train': os.path.join(base_path, "Labels/train"),
'val': os.path.join(base_path, "Labels/val"),
'test': os.path.join(base_path, "Labels/test")
}

Ensure annotations exist

def check_annotations(label_dirs):
for split, dir_path in label_dirs.items():
if not os.listdir(dir_path):
raise ValueError(f"No annotations found in {dir_path}")

check_annotations(label_dirs)

Create cache files in the correct subdirectories

def create_cache_files(image_dirs, label_dirs):
for split, image_dir in image_dirs.items():
cache_file = os.path.join(image_dir, f'{split}.cache')
if not os.path.exists(cache_file):
with open(cache_file, 'w') as f:
f.write('')

    label_dir = label_dirs[split]
    cache_file = os.path.join(label_dir, f'{split}.cache')
    if not os.path.exists(cache_file):
        with open(cache_file, 'w') as f:
            f.write('')

create_cache_files(image_dirs, label_dirs)

Train YOLOv8 model

model = YOLO('yolov8s.pt')
model.train(data=os.path.join(base_path, 'dataset.yaml'), epochs=50, imgsz=640, batch=16, cache=True, name='poker_cards6')

PowerShell Command

Navigate to the project directory

cd C:\Users\Louis\Desktop\Bot

Activate the virtual environment

.\venv\Scripts\Activate.ps1

Install dependencies

pip install -r requirements.txt

Execute the training script

python -c "
import os
from ultralytics import YOLO

Define paths

base_path = 'C:/Users/Louis/Desktop/Bot/dataset'
image_dirs = {
'train': os.path.join(base_path, 'Images/train'),
'val': os.path.join(base_path, 'Images/val'),
'test': os.path.join(base_path, 'Images/test')
}

label_dirs = {
'train': os.path.join(base_path, 'Labels/train'),
'val': os.path.join(base_path, 'Labels/val'),
'test': os.path.join(base_path, 'Labels/test')
}

Ensure annotations exist

def check_annotations(label_dirs):
for split, dir_path in label_dirs.items():
if not os.listdir(dir_path):
raise ValueError(f'No annotations found in {dir_path}')

check_annotations(label_dirs)

Create cache files in the correct subdirectories

def create_cache_files(image_dirs, label_dirs):
for split, image_dir in image_dirs.items():
cache_file = os.path.join(image_dir, f'{split}.cache')
if not os.path.exists(cache_file):
with open(cache_file, 'w') as f:
f.write('')

    label_dir = label_dirs[split]
    cache_file = os.path.join(label_dir, f'{split}.cache')
    if not os.path.exists(cache_file):
        with open(cache_file, 'w') as f:
            f.write('')

create_cache_files(image_dirs, label_dirs)

Train YOLOv8 model

model = YOLO('yolov8s.pt')
model.train(data=os.path.join(base_path, 'dataset.yaml'), epochs=50, imgsz=640, batch=16, cache=True, name='poker_cards6')
"
Thank you all for taking the time to read through my issue and for any assistance you can provide. Your help is greatly appreciated, and I look forward to resolving this with your expert guidance.

[pderrenger]

@aeromista2a hi Louis,

Thank you for providing such a detailed description of your issue and the steps you've taken so far. It sounds like you've been thorough in setting up your dataset and training script. Let's try to address the problem with the cache files not being utilized correctly.

Steps to Troubleshoot

1. Verify Dataset Structure and Paths:
   Ensure that the paths in your dataset.yaml file are correct and point to the appropriate directories. It looks like your structure is correct, but double-checking won't hurt.

   path: C:/Users/Louis/Desktop/Bot/dataset
   train: 
     images: Images/train
     labels: Labels/train
   val: 
     images: Images/val
     labels: Labels/val
   test: 
     images: Images/test
     labels: Labels/test
   
   nc: 137
   
   names:
     0: a_coeur
     1: 2_coeur
     ...
     136: tapis j1 table de 8

2. Check Cache File Creation:
   Ensure that the cache files are being created correctly and contain the necessary data. The cache files should not be empty. You can manually inspect these files to ensure they are populated with the correct information.

3. Update to Latest Version:
   Make sure you are using the latest version of the Ultralytics YOLO package. Sometimes, issues are resolved in newer releases. You can update the package using:

   pip install --upgrade ultralytics

4. Recreate Cache Files:
   If the cache files are not being utilized correctly, try deleting the existing cache files and let the training script recreate them. This can sometimes resolve issues with corrupted or improperly formatted cache files.

5. Use Absolute Paths:
   Ensure that the paths in your dataset.yaml and training script are absolute paths. Relative paths can sometimes cause issues, especially in complex directory structures.

Example Code to Recreate Cache Files

Here's a modified version of your script to ensure cache files are created correctly:

import os
from ultralytics import YOLO

# Define paths
base_path = 'C:/Users/Louis/Desktop/Bot/dataset'
image_dirs = {
    'train': os.path.join(base_path, 'Images/train'),
    'val': os.path.join(base_path, 'Images/val'),
    'test': os.path.join(base_path, 'Images/test')
}

label_dirs = {
    'train': os.path.join(base_path, 'Labels/train'),
    'val': os.path.join(base_path, 'Labels/val'),
    'test': os.path.join(base_path, 'Labels/test')
}

# Ensure annotations exist
def check_annotations(label_dirs):
    for split, dir_path in label_dirs.items():
        if not os.listdir(dir_path):
            raise ValueError(f'No annotations found in {dir_path}')

check_annotations(label_dirs)

# Create cache files in the correct subdirectories
def create_cache_files(image_dirs, label_dirs):
    for split, image_dir in image_dirs.items():
        cache_file = os.path.join(image_dir, f'{split}.cache')
        if not os.path.exists(cache_file):
            with open(cache_file, 'w') as f:
                f.write('')

        label_dir = label_dirs[split]
        cache_file = os.path.join(label_dir, f'{split}.cache')
        if not os.path.exists(cache_file):
            with open(cache_file, 'w') as f:
                f.write('')

create_cache_files(image_dirs, label_dirs)

# Train YOLOv8 model
model = YOLO('yolov8s.pt')
model.train(data=os.path.join(base_path, 'dataset.yaml'), epochs=50, imgsz=640, batch=16, cache=True, name='poker_cards6')

Additional Resources

For more detailed guidance on dataset formatting and troubleshooting, you can refer to the Ultralytics Dataset Formatting Guide and the Common Issues Guide.

If the issue persists, please provide a minimum reproducible example as outlined here. This will help us diagnose the problem more effectively.

I hope this helps! Please let us know if you have any further questions or if the issue persists.