YOLOv8 Image Classifier

Using YOLOv8 to classify an entire image into one of a set of predefined classes.

• Github Repository

• YOLOv8 Image Classifier
  • Docker Environment
  • Training Results
  • Model Export Options
  • YOLOv8n
    • Model Evaluation
    • Model Predictions
    • Model Export
  • YOLOv8s
    • Model Evaluation
    • Model Predictions
    • Model Export
  • YOLOv8m
    • Model Evaluation
    • Model Predictions
    • Model Export
  • YOLOv8l
    • Model Evaluation
    • Model Predictions
    • Model Export
  • YOLOv8x
    • Model Evaluation
    • Model Export

Model	size (pixels)	acc top1	acc top5	Speed CPU ONNX (ms)	Speed A100 TensorRT (ms)	params (M)	FLOPs (B) at 640
YOLOv8n-cls	224	66.6	87.0	12.9	0.31	2.7	4.3
YOLOv8s-cls	224	72.3	91.1	23.4	0.35	6.4	13.5
YOLOv8m-cls	224	76.4	93.2	85.4	0.62	17.0	42.7
YOLOv8l-cls	224	78.0	94.1	163.0	0.87	37.5	99.7
YOLOv8x-cls	224	78.4	94.3	232.0	1.01	57.4	154.8

Docker Environment

• PyTorch Jupyter

docker run --gpus all -ti --rm \
    -v $(pwd):/opt/app -p 8888:8888 \
    --name pytorch-jupyter \
    pytorch-jupyter:latest

Training Results

• Flower Dataset (12,297 files / 48 classes)

	top1_acc	top5_acc	validation [it/s]	inference [ms]	prediction preprocess [ms]	prediction inference [ms]	prediction postprocess [ms]	weights [MB]	weights (ONNX) [MB]	weights (TensorRT) [MB]
YOLOv8n	0.792	0.954	86.50	0.2	3.600000	2.733333	0.033333	2.935699	5.731391	7.929402
YOLOv8s	0.832	0.965	91.72	0.2	5.333333	4.500000	0.100000	9.890900	19.618174	26.030774
YOLOv8m	0.858	0.969	87.90	0.4	0.466667	5.000000	0.066667	30.324455	60.397784	63.700307
YOLOv8l	0.850	0.973	72.36	0.8	0.500000	7.300000	0.066667	69.331619	138.311538	166.345273
YOLOv8x	0.861	0.971	53.22	1.1	2.166667	14.400000	0.100000	107.382278	214.371352	225.140141

Model Export Options

Format	format Argument	Model	Metadata	Arguments
PyTorch	-	yolov8n-cls.pt	✅	-
TorchScript	torchscript	yolov8n-cls.torchscript	✅	imgsz, optimize
ONNX	onnx	yolov8n-cls.onnx	✅	imgsz, half, dynamic, simplify, opset
OpenVINO	openvino	yolov8n-cls_openvino_model/	✅	imgsz, half
TensorRT	engine	yolov8n-cls.engine	✅	imgsz, half, dynamic, simplify, workspace
CoreML	coreml	yolov8n-cls.mlpackage	✅	imgsz, half, int8, nms
TF SavedModel	saved_model	yolov8n-cls_saved_model/	✅	imgsz, keras
TF GraphDef	pb	yolov8n-cls.pb	❌	imgsz
TF Lite	tflite	yolov8n-cls.tflite	✅	imgsz, half, int8
TF Edge TPU	edgetpu	yolov8n-cls_edgetpu.tflite	✅	imgsz
TF.js	tfjs	yolov8n-cls_web_model/	✅	imgsz
PaddlePaddle	paddle	yolov8n-cls_paddle_model/	✅	imgsz
ncnn	ncnn	yolov8n-cls_ncnn_model/	✅	imgsz, half

!pip install nvidia-tensorrt sng4onnx onnx_graphsurgeon onnx onnxsim onnxruntime-gpu # tensorflow tflite_support onnx2tf

from ultralytics import YOLO

Model	size (pixels)	acc top1	acc top5	Speed CPU ONNX (ms)	Speed A100 TensorRT (ms)	params (M)	FLOPs (B) at 640
YOLOv8n-cls	224	66.6	87.0	12.9	0.31	2.7	4.3
YOLOv8s-cls	224	72.3	91.1	23.4	0.35	6.4	13.5
YOLOv8m-cls	224	76.4	93.2	85.4	0.62	17.0	42.7
YOLOv8l-cls	224	78.0	94.1	163.0	0.87	37.5	99.7
YOLOv8x-cls	224	78.4	94.3	232.0	1.01	57.4	154.8

YOLOv8n

model = YOLO('yolov8n-cls.pt')  # load a pretrained model (recommended for training)

# Train the model
results = model.train(data='./data/Flower_Dataset', epochs=20, imgsz=64)

Model Evaluation

# Load a model
model_n = YOLO('./runs/classify/train_yolov8n/weights/last.pt')  # load a custom model

# Validate the model
metrics_n = model_n.val() # no arguments needed, dataset and settings remembered
print(metrics_n.top1) # top1 accuracy: 0.7915857434272766
print(metrics_n.top5) # top5 accuracy: 0.9543688893318176

Model Predictions

# Predict with the model
results_n = model_n('./assets/snapshots/Viola_Tricolor.jpg')

# image 1/1 /opt/app/assets/snapshots/Viola_Tricolor.jpg: 64x64 Viola 1.00, Aquilegia 0.00, Malvaceae 0.00, Helianthus_Annuus 0.00, Plumeria 0.00, 3.6ms
# Speed: 0.6ms preprocess, 3.6ms inference, 0.1ms postprocess per image at shape (1, 3, 64, 64)

results_n = model_n('./assets/snapshots/Strelitzia.jpg')

# image 1/1 /opt/app/assets/snapshots/Strelitzia.jpg: 64x64 Strelitzia_Reginae 1.00, Alstroemeria 0.00, Guzmania 0.00, Rose 0.00, Anthurium_Andraeanum 0.00, 2.5ms
# Speed: 7.1ms preprocess, 2.5ms inference, 0.0ms postprocess per image at shape (1, 3, 64, 64)

results_n = model_n('./assets/snapshots/Water_Lilly.jpg')

# image 1/1 /opt/app/assets/snapshots/Water_Lilly.jpg: 64x64 Nymphaea_Tetragona 0.78, Dahlia 0.20, Rose 0.01, Alstroemeria 0.01, Antirrhinum 0.00, 3.1ms
# Speed: 6.6ms preprocess, 2.1ms inference, 0.0ms postprocess per image at shape (1, 3, 64, 64)

Model Export

Format	format Argument	Model	Metadata	Arguments
PyTorch	-	yolov8n-cls.pt	✅	-
TorchScript	torchscript	yolov8n-cls.torchscript	✅	imgsz, optimize
ONNX	onnx	yolov8n-cls.onnx	✅	imgsz, half, dynamic, simplify, opset
OpenVINO	openvino	yolov8n-cls_openvino_model/	✅	imgsz, half
TensorRT	engine	yolov8n-cls.engine	✅	imgsz, half, dynamic, simplify, workspace
CoreML	coreml	yolov8n-cls.mlpackage	✅	imgsz, half, int8, nms
TF SavedModel	saved_model	yolov8n-cls_saved_model/	✅	imgsz, keras
TF GraphDef	pb	yolov8n-cls.pb	❌	imgsz
TF Lite	tflite	yolov8n-cls.tflite	✅	imgsz, half, int8
TF Edge TPU	edgetpu	yolov8n-cls_edgetpu.tflite	✅	imgsz
TF.js	tfjs	yolov8n-cls_web_model/	✅	imgsz
PaddlePaddle	paddle	yolov8n-cls_paddle_model/	✅	imgsz
ncnn	ncnn	yolov8n-cls_ncnn_model/	✅	imgsz, half

model_n.export(format='onnx')
model_n.export(format='engine')

YOLOv8s

model = YOLO('yolov8s-cls.pt')

results = model.train(data='./data/Flower_Dataset', epochs=20, imgsz=64)

Model Evaluation

# Load a model
model_s = YOLO('./runs/classify/train_yolov8s/weights/last.pt')  # load a custom model

# Validate the model
metrics_s = model_s.val()  # no arguments needed, dataset and settings remembered
print(metrics_s.top1)   # top1 accuracy: 0.8323624134063721
print(metrics_s.top5)   # top5 accuracy: 0.9650484919548035

Model Predictions

# Predict with the model
results_s = model_s('./assets/snapshots/Viola_Tricolor.jpg')

# image 1/1 /opt/app/assets/snapshots/Viola_Tricolor.jpg: 64x64 Viola 1.00, Primula 0.00, Malvaceae 0.00, Plumeria 0.00, Datura_Metel 0.00, 2.7ms
# Speed: 0.5ms preprocess, 2.7ms inference, 0.1ms postprocess per image at shape (1, 3, 64, 64)

results_s = model_s('./assets/snapshots/Strelitzia.jpg')

# image 1/1 /opt/app/assets/snapshots/Strelitzia.jpg: 64x64 Strelitzia_Reginae 1.00, Helianthus_Annuus 0.00, Nymphaea_Tetragona 0.00, Plumeria 0.00, Crocus 0.00, 7.6ms
# Speed: 15.1ms preprocess, 7.6ms inference, 0.1ms postprocess per image at shape (1, 3, 64, 64)

results_s = model_s('./assets/snapshots/Water_Lilly.jpg')

# image 1/1 /opt/app/assets/snapshots/Water_Lilly.jpg: 64x64 Nymphaea_Tetragona 0.99, Alstroemeria 0.01, Passiflora 0.00, Billbergia_Pyramidalis 0.00, Protea_Cynaroides 0.00, 3.2ms
# Speed: 0.4ms preprocess, 3.2ms inference, 0.1ms postprocess per image at shape (1, 3, 64, 64)

Model Export

model_s.export(format='onnx')
model_s.export(format='engine')

YOLOv8m

model = YOLO('yolov8m-cls.pt')

results = model.train(data='./data/Flower_Dataset', epochs=20, imgsz=64)

Model Evaluation

# Load a model
model_m = YOLO('./runs/classify/train_yolov8m/weights/last.pt')  # load a custom model

# Validate the model
metrics_m = model_m.val()  # no arguments needed, dataset and settings remembered
print(metrics_m.top1)   # top1 accuracy: 0.8579287528991699
print(metrics_m.top5)   # top5 accuracy: 0.9692556262016296

Model Predictions

# Predict with the model
results_m = model_m('./assets/snapshots/Viola_Tricolor.jpg')

# image 1/1 /opt/app/assets/snapshots/Viola_Tricolor.jpg: 64x64 Viola 1.00, Iris_Pseudacorus 0.00, Primula 0.00, Cattleya 0.00, Helianthus_Annuus 0.00, 4.0ms
# Speed: 0.4ms preprocess, 4.0ms inference, 0.0ms postprocess per image at shape (1, 3, 64, 64)

results_m = model_m('./assets/snapshots/Strelitzia.jpg')

# image 1/1 /opt/app/assets/snapshots/Strelitzia.jpg: 64x64 Strelitzia_Reginae 1.00, Alpinia_Purpurata 0.00, Zantedeschia_Aethiopica 0.00, Helianthus_Annuus 0.00, Plumeria 0.00, 6.0ms
# Speed: 0.5ms preprocess, 6.0ms inference, 0.1ms postprocess per image at shape (1, 3, 64, 64)

results_m = model_m('./assets/snapshots/Water_Lilly.jpg')

# image 1/1 /opt/app/assets/snapshots/Water_Lilly.jpg: 64x64 Nymphaea_Tetragona 1.00, Dahlia 0.00, Alstroemeria 0.00, Passiflora 0.00, Guzmania 0.00, 5.0ms
# Speed: 0.5ms preprocess, 5.0ms inference, 0.1ms postprocess per image at shape (1, 3, 64, 64)

Model Export

model_m.export(format='onnx')
model_m.export(format='engine')

YOLOv8l

model = YOLO('yolov8l-cls.pt')

results = model.train(data='./data/Flower_Dataset', epochs=20, imgsz=64)

Model Evaluation

# Load a model
model_l = YOLO('./runs/classify/train_yolov8l/weights/last.pt')  # load a custom model

# Validate the model
metrics_l = model_l.val()  # no arguments needed, dataset and settings remembered
print(metrics_l.top1)   # top1 accuracy: 0.849838137626648
print(metrics_l.top5)   # top5 accuracy: 0.9731391072273254

Model Predictions

# Predict with the model
results_l = model_l('./assets/snapshots/Viola_Tricolor.jpg')

# image 1/1 /opt/app/assets/snapshots/Viola_Tricolor.jpg: 64x64 Viola 1.00, Malvaceae 0.00, Iris_Pseudacorus 0.00, Primula 0.00, Paphiopedilum 0.00, 6.3ms
# Speed: 0.5ms preprocess, 6.3ms inference, 0.1ms postprocess per image at shape (1, 3, 64, 64)

results_l = model_l('./assets/snapshots/Strelitzia.jpg')

# image 1/1 /opt/app/assets/snapshots/Strelitzia.jpg: 64x64 Strelitzia_Reginae 1.00, Anthurium_Andraeanum 0.00, Tropaeolum_Majus 0.00, Alstroemeria 0.00, Zantedeschia_Aethiopica 0.00, 8.0ms
# Speed: 0.6ms preprocess, 8.0ms inference, 0.0ms postprocess per image at shape (1, 3, 64, 64)

results_l = model_l('./assets/snapshots/Water_Lilly.jpg')

# image 1/1 /opt/app/assets/snapshots/Water_Lilly.jpg: 64x64 Nymphaea_Tetragona 1.00, Rose 0.00, Primula 0.00, Dahlia 0.00, Alstroemeria 0.00, 7.6ms
# Speed: 0.4ms preprocess, 7.6ms inference, 0.1ms postprocess per image at shape (1, 3, 64, 64)

Model Export

model_l.export(format='onnx')
model_l.export(format='engine')

YOLOv8x

model = YOLO('yolov8x-cls.pt')

results = model.train(data='./data/Flower_Dataset', epochs=20, imgsz=64)

Model Evaluation

# Load a model
model_x = YOLO('./runs/classify/train_yolov8x/weights/last.pt')  # load a custom model

# Validate the model
metrics_x = model_x.val()  # no arguments needed, dataset and settings remembered
print(metrics_x.top1)   # top1 accuracy: 0.8605177402496338
print(metrics_x.top5)   # top5 accuracy: 0.9708737730979919


![YOLOv8 Image Classifier](./model_x_confusion_matrix_normalized.webp)

### Model Predictions

```python
# Predict with the model
results_x = model_x('./assets/snapshots/Viola_Tricolor.jpg')

# image 1/1 /opt/app/assets/snapshots/Viola_Tricolor.jpg: 64x64 Viola 1.00, Dahlia 0.00, Tropaeolum_Majus 0.00, Rose 0.00, Primula 0.00, 10.1ms
# Speed: 0.5ms preprocess, 10.1ms inference, 0.1ms postprocess per image at shape (1, 3, 64, 64)

results_x = model_x('./assets/snapshots/Strelitzia.jpg')

# image 1/1 /opt/app/assets/snapshots/Strelitzia.jpg: 64x64 Strelitzia_Reginae 1.00, Tropaeolum_Majus 0.00, Alstroemeria 0.00, Rose 0.00, Helianthus_Annuus 0.00, 24.0ms
# Speed: 5.5ms preprocess, 24.0ms inference, 0.1ms postprocess per image at shape (1, 3, 64, 64)

results_x = model_x('./assets/snapshots/Water_Lilly.jpg')

# image 1/1 /opt/app/assets/snapshots/Water_Lilly.jpg: 64x64 Nymphaea_Tetragona 0.99, Rose 0.00, Alstroemeria 0.00, Primula 0.00, Protea_Cynaroides 0.00, 9.1ms
# Speed: 0.5ms preprocess, 9.1ms inference, 0.1ms postprocess per image at shape (1, 3, 64, 64)

Model Export

model_x.export(format='onnx')
model_x.export(format='engine')