mirrors
/
PaddleClas
mirror of https://github.com/PaddlePaddle/PaddleClas.git
1
0
Fork 0
Code Issues Projects Releases Wiki Activity

update config file for multi-label image classification

pull/3232/head
zhouchangda 2024-08-29 03:56:40 +00:00 committed by Tingquan Gao
parent bc9401f861
commit 9c37234cfb
7 changed files with 1015 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

172
ppcls/configs/MultiLabelCOCO/MLDecoder/CLIP_vit_base_patch16_448_ml_decoder_448.yaml 100644
View File

@ -0,0 +1,172 @@
# global configs
Global:
checkpoints: null
pretrained_model: null
output_dir: ./output/
device: gpu
save_interval: 10
eval_during_train: True
eval_interval: 1
epochs: 40
print_batch_step: 10
use_visualdl: False
# used for static mode and model export
image_shape: [3, 448, 448]
save_inference_dir: ./inference
# training model under @to_static
to_static: False
use_multilabel: True
# model ema
EMA:
decay: 0.9997
# mixed precision
AMP:
use_amp: False
use_fp16_test: False
scale_loss: 128.0
use_dynamic_loss_scaling: True
use_promote: False
# O1: mixed fp16, O2: pure fp16
level: O2
# model architecture
Arch:
name: CLIP_vit_base_patch16_224
class_num: 80
return_embed: False
use_fused_attn: False # fused attn can be used in AMP O2 mode only
pretrained: True
use_ml_decoder: True
# ml-decoder head
MLDecoder:
query_num: 80 # default: 80, query_num <= class_num
in_channels: 768
remove_layers: []
replace_layer: 'head'
# loss function config for training/eval process
Loss:
Train:
- MultiLabelAsymmetricLoss:
weight: 1.0
gamma_pos: 0
gamma_neg: 4
clip: 0.05
disable_focal_loss_grad: True
Eval:
- MultiLabelAsymmetricLoss:
weight: 1.0
gamma_pos: 0
gamma_neg: 4
clip: 0.05
disable_focal_loss_grad: True
Optimizer:
name: AdamW
beta1: 0.9
beta2: 0.999
epsilon: 1e-8
weight_decay: 1e-4
one_dim_param_no_weight_decay: True
lr:
name: Cosine
learning_rate: 5e-5
eta_min: 1e-10
warmup_epoch: 5
warmup_start_lr: 1e-6
# data loader for train and eval
DataLoader:
Train:
dataset:
name: MultiLabelDataset
image_root: dataset/coco_ml/images
cls_label_path: dataset/coco_ml/train.txt
transform_ops:
- DecodeImage:
to_rgb: True
channel_first: False
- ResizeImage:
size: 448
interpolation: bilinear
backend: pil
- Cutout:
length: 224
fill_value: none
- RandAugmentV4:
- NormalizeImage:
scale: 1.0/255.0
mean: [0.485, 0.456, 0.406]
std: [0.229, 0.224, 0.225]
order: ''
sampler:
name: DistributedBatchSampler
batch_size: 16
drop_last: False
shuffle: True
loader:
num_workers: 8
use_shared_memory: True
Eval:
dataset:
name: MultiLabelDataset
image_root: dataset/coco_ml/images
cls_label_path: dataset/coco_ml/val.txt
transform_ops:
- DecodeImage:
to_rgb: True
channel_first: False
- ResizeImage:
size: 448
interpolation: bilinear
backend: pil
- NormalizeImage:
scale: 1.0/255.0
mean: [0.485, 0.456, 0.406]
std: [0.229, 0.224, 0.225]
order: ''
sampler:
name: DistributedBatchSampler
batch_size: 8
drop_last: False
shuffle: False
loader:
num_workers: 8
use_shared_memory: True
Infer:
infer_imgs: deploy/images/coco_000000570688.jpg
batch_size: 10
transforms:
- DecodeImage:
to_rgb: True
channel_first: False
- ResizeImage:
size: 448
interpolation: bilinear
backend: pil
- NormalizeImage:
scale: 1.0/255.0
mean: [0.485, 0.456, 0.406]
std: [0.229, 0.224, 0.225]
order: ''
- ToCHWImage:
PostProcess:
name: MultiLabelThreshOutput
threshold: 0.5
class_id_map_file: dataset/coco_ml/label.txt
Metric:
Train:
Eval:
- MultiLabelMAP:
# support list: integral, 11point
# default: integral
map_type: integral

168
ppcls/configs/MultiLabelCOCO/MLDecoder/PP-HGNetV2-B0_ml_decoder_448.yaml 100644
View File

@ -0,0 +1,168 @@
# global configs
Global:
checkpoints: null
pretrained_model: null
output_dir: ./output/
device: gpu
save_interval: 10
eval_during_train: True
eval_interval: 1
epochs: 40
print_batch_step: 10
use_visualdl: False
# used for static mode and model export
image_shape: [3, 448, 448]
save_inference_dir: ./inference
# training model under @to_static
to_static: False
use_multilabel: True
# model ema
EMA:
decay: 0.9997
# mixed precision
AMP:
use_amp: True
use_fp16_test: False
scale_loss: 128.0
use_dynamic_loss_scaling: True
use_promote: False
# O1: mixed fp16, O2: pure fp16
level: O1
# model architecture
Arch:
name: PPHGNetV2_B0
class_num: 80
pretrained: True # ssld pretrained
use_ml_decoder: True
# ml-decoder head
MLDecoder:
query_num: 80 # default: 80, query_num <= class_num
in_channels: 2048
# loss function config for training/eval process
Loss:
Train:
- MultiLabelAsymmetricLoss:
weight: 1.0
gamma_pos: 0
gamma_neg: 4
clip: 0.05
disable_focal_loss_grad: True
Eval:
- MultiLabelAsymmetricLoss:
weight: 1.0
gamma_pos: 0
gamma_neg: 4
clip: 0.05
disable_focal_loss_grad: True
Optimizer:
name: AdamW
beta1: 0.9
beta2: 0.999
epsilon: 1e-8
weight_decay: 1e-4
one_dim_param_no_weight_decay: True
lr:
name: Cosine
learning_rate: 1e-4
eta_min: 1e-10
warmup_epoch: 5
warmup_start_lr: 1e-6
# data loader for train and eval
DataLoader:
Train:
dataset:
name: MultiLabelDataset
image_root: dataset/coco_ml/images
cls_label_path: dataset/coco_ml/train.txt
transform_ops:
- DecodeImage:
to_rgb: True
channel_first: False
- ResizeImage:
size: 448
interpolation: bilinear
backend: pil
- Cutout:
length: 224
fill_value: none
- RandAugmentV4:
- NormalizeImage:
scale: 1.0/255.0
mean: [0.485, 0.456, 0.406]
std: [0.229, 0.224, 0.225]
order: ''
sampler:
name: DistributedBatchSampler
batch_size: 64
drop_last: False
shuffle: True
loader:
num_workers: 8
use_shared_memory: True
Eval:
dataset:
name: MultiLabelDataset
image_root: dataset/coco_ml/images
cls_label_path: dataset/coco_ml/val.txt
transform_ops:
- DecodeImage:
to_rgb: True
channel_first: False
- ResizeImage:
size: 448
interpolation: bilinear
backend: pil
- NormalizeImage:
scale: 1.0/255.0
mean: [0.485, 0.456, 0.406]
std: [0.229, 0.224, 0.225]
order: ''
sampler:
name: DistributedBatchSampler
batch_size: 32
drop_last: False
shuffle: False
loader:
num_workers: 8
use_shared_memory: True
Infer:
infer_imgs: deploy/images/coco_000000570688.jpg
batch_size: 10
transforms:
- DecodeImage:
to_rgb: True
channel_first: False
- ResizeImage:
size: 448
interpolation: bilinear
backend: pil
- NormalizeImage:
scale: 1.0/255.0
mean: [0.485, 0.456, 0.406]
std: [0.229, 0.224, 0.225]
order: ''
- ToCHWImage:
PostProcess:
name: MultiLabelThreshOutput
threshold: 0.5
class_id_map_file: dataset/coco_ml/label.txt
Metric:
Train:
Eval:
- MultiLabelMAP:
# support list: integral, 11point
# default: integral
map_type: integral

168
ppcls/configs/MultiLabelCOCO/MLDecoder/PP-HGNetV2-B4_ml_decoder_448.yaml 100644
View File

@ -0,0 +1,168 @@
# global configs
Global:
checkpoints: null
pretrained_model: null
output_dir: ./output/
device: gpu
save_interval: 10
eval_during_train: True
eval_interval: 1
epochs: 40
print_batch_step: 10
use_visualdl: False
# used for static mode and model export
image_shape: [3, 448, 448]
save_inference_dir: ./inference
# training model under @to_static
to_static: False
use_multilabel: True
# model ema
EMA:
decay: 0.9997
# mixed precision
AMP:
use_amp: True
use_fp16_test: False
scale_loss: 128.0
use_dynamic_loss_scaling: True
use_promote: False
# O1: mixed fp16, O2: pure fp16
level: O1
# model architecture
Arch:
name: PPHGNetV2_B4
class_num: 80
pretrained: True # ssld pretrained
use_ml_decoder: True
# ml-decoder head
MLDecoder:
query_num: 80 # default: 80, query_num <= class_num
in_channels: 2048
# loss function config for training/eval process
Loss:
Train:
- MultiLabelAsymmetricLoss:
weight: 1.0
gamma_pos: 0
gamma_neg: 4
clip: 0.05
disable_focal_loss_grad: True
Eval:
- MultiLabelAsymmetricLoss:
weight: 1.0
gamma_pos: 0
gamma_neg: 4
clip: 0.05
disable_focal_loss_grad: True
Optimizer:
name: AdamW
beta1: 0.9
beta2: 0.999
epsilon: 1e-8
weight_decay: 1e-4
one_dim_param_no_weight_decay: True
lr:
name: Cosine
learning_rate: 1e-4
eta_min: 1e-10
warmup_epoch: 5
warmup_start_lr: 1e-6
# data loader for train and eval
DataLoader:
Train:
dataset:
name: MultiLabelDataset
image_root: dataset/coco_ml/images
cls_label_path: dataset/coco_ml/train.txt
transform_ops:
- DecodeImage:
to_rgb: True
channel_first: False
- ResizeImage:
size: 448
interpolation: bilinear
backend: pil
- Cutout:
length: 224
fill_value: none
- RandAugmentV4:
- NormalizeImage:
scale: 1.0/255.0
mean: [0.485, 0.456, 0.406]
std: [0.229, 0.224, 0.225]
order: ''
sampler:
name: DistributedBatchSampler
batch_size: 64
drop_last: False
shuffle: True
loader:
num_workers: 8
use_shared_memory: True
Eval:
dataset:
name: MultiLabelDataset
image_root: dataset/coco_ml/images
cls_label_path: dataset/coco_ml/val.txt
transform_ops:
- DecodeImage:
to_rgb: True
channel_first: False
- ResizeImage:
size: 448
interpolation: bilinear
backend: pil
- NormalizeImage:
scale: 1.0/255.0
mean: [0.485, 0.456, 0.406]
std: [0.229, 0.224, 0.225]
order: ''
sampler:
name: DistributedBatchSampler
batch_size: 16
drop_last: False
shuffle: False
loader:
num_workers: 8
use_shared_memory: True
Infer:
infer_imgs: deploy/images/coco_000000570688.jpg
batch_size: 10
transforms:
- DecodeImage:
to_rgb: True
channel_first: False
- ResizeImage:
size: 448
interpolation: bilinear
backend: pil
- NormalizeImage:
scale: 1.0/255.0
mean: [0.485, 0.456, 0.406]
std: [0.229, 0.224, 0.225]
order: ''
- ToCHWImage:
PostProcess:
name: MultiLabelThreshOutput
threshold: 0.5
class_id_map_file: dataset/coco_ml/label.txt
Metric:
Train:
Eval:
- MultiLabelMAP:
# support list: integral, 11point
# default: integral
map_type: integral

168
ppcls/configs/MultiLabelCOCO/MLDecoder/PP-HGNetV2-B6_ml_decoder_448.yaml 100644
View File

@ -0,0 +1,168 @@
# global configs
Global:
checkpoints: null
pretrained_model: null
output_dir: ./output/
device: gpu
save_interval: 10
eval_during_train: True
eval_interval: 1
epochs: 40
print_batch_step: 10
use_visualdl: False
# used for static mode and model export
image_shape: [3, 448, 448]
save_inference_dir: ./inference
# training model under @to_static
to_static: False
use_multilabel: True
# model ema
EMA:
decay: 0.9997
# mixed precision
AMP:
use_amp: True
use_fp16_test: False
scale_loss: 128.0
use_dynamic_loss_scaling: True
use_promote: False
# O1: mixed fp16, O2: pure fp16
level: O1
# model architecture
Arch:
name: PPHGNetV2_B6
class_num: 80
pretrained: True # ssld pretrained
use_ml_decoder: True
# ml-decoder head
MLDecoder:
query_num: 80 # default: 80, query_num <= class_num
in_channels: 2048
# loss function config for training/eval process
Loss:
Train:
- MultiLabelAsymmetricLoss:
weight: 1.0
gamma_pos: 0
gamma_neg: 4
clip: 0.05
disable_focal_loss_grad: True
Eval:
- MultiLabelAsymmetricLoss:
weight: 1.0
gamma_pos: 0
gamma_neg: 4
clip: 0.05
disable_focal_loss_grad: True
Optimizer:
name: AdamW
beta1: 0.9
beta2: 0.999
epsilon: 1e-8
weight_decay: 1e-4
one_dim_param_no_weight_decay: True
lr:
name: Cosine
learning_rate: 1e-4
eta_min: 1e-10
warmup_epoch: 5
warmup_start_lr: 1e-6
# data loader for train and eval
DataLoader:
Train:
dataset:
name: MultiLabelDataset
image_root: dataset/coco_ml/images
cls_label_path: dataset/coco_ml/train.txt
transform_ops:
- DecodeImage:
to_rgb: True
channel_first: False
- ResizeImage:
size: 448
interpolation: bilinear
backend: pil
- Cutout:
length: 224
fill_value: none
- RandAugmentV4:
- NormalizeImage:
scale: 1.0/255.0
mean: [0.485, 0.456, 0.406]
std: [0.229, 0.224, 0.225]
order: ''
sampler:
name: DistributedBatchSampler
batch_size: 32
drop_last: False
shuffle: True
loader:
num_workers: 8
use_shared_memory: True
Eval:
dataset:
name: MultiLabelDataset
image_root: dataset/coco_ml/images
cls_label_path: dataset/coco_ml/val.txt
transform_ops:
- DecodeImage:
to_rgb: True
channel_first: False
- ResizeImage:
size: 448
interpolation: bilinear
backend: pil
- NormalizeImage:
scale: 1.0/255.0
mean: [0.485, 0.456, 0.406]
std: [0.229, 0.224, 0.225]
order: ''
sampler:
name: DistributedBatchSampler
batch_size: 8
drop_last: False
shuffle: False
loader:
num_workers: 8
use_shared_memory: True
Infer:
infer_imgs: deploy/images/coco_000000570688.jpg
batch_size: 10
transforms:
- DecodeImage:
to_rgb: True
channel_first: False
- ResizeImage:
size: 448
interpolation: bilinear
backend: pil
- NormalizeImage:
scale: 1.0/255.0
mean: [0.485, 0.456, 0.406]
std: [0.229, 0.224, 0.225]
order: ''
- ToCHWImage:
PostProcess:
name: MultiLabelThreshOutput
threshold: 0.5
class_id_map_file: dataset/coco_ml/label.txt
Metric:
Train:
Eval:
- MultiLabelMAP:
# support list: integral, 11point
# default: integral
map_type: integral

170
ppcls/configs/MultiLabelCOCO/MLDecoder/PP-LCNet_x1_0_ml_decoder_448.yaml 100644
View File

@ -0,0 +1,170 @@
# global configs
Global:
checkpoints: null
pretrained_model: null
output_dir: ./output/
device: gpu
save_interval: 10
eval_during_train: True
eval_interval: 1
epochs: 40
print_batch_step: 10
use_visualdl: False
# used for static mode and model export
image_shape: [3, 448, 448]
save_inference_dir: ./inference
# training model under @to_static
to_static: False
use_multilabel: True
# model ema
EMA:
decay: 0.9997
# mixed precision
AMP:
use_amp: True
use_fp16_test: False
scale_loss: 128.0
use_dynamic_loss_scaling: True
use_promote: False
# O1: mixed fp16, O2: pure fp16
level: O1
# model architecture
Arch:
name: PPLCNet_x1_0
class_num: 80
pretrained: True
use_ml_decoder: True
# ml-decoder head
MLDecoder:
query_num: 80 # default: 80, query_num <= class_num
class_num: 80
in_channels: 1280
# loss function config for training/eval process
Loss:
Train:
- MultiLabelAsymmetricLoss:
weight: 1.0
gamma_pos: 0
gamma_neg: 4
clip: 0.05
disable_focal_loss_grad: True
Eval:
- MultiLabelAsymmetricLoss:
weight: 1.0
gamma_pos: 0
gamma_neg: 4
clip: 0.05
disable_focal_loss_grad: True
Optimizer:
name: AdamW
beta1: 0.9
beta2: 0.999
epsilon: 1e-8
weight_decay: 1e-4
one_dim_param_no_weight_decay: True
lr:
name: Cosine
learning_rate: 1e-4
eta_min: 1e-10
warmup_epoch: 5
warmup_start_lr: 1e-6
# data loader for train and eval
DataLoader:
Train:
dataset:
name: MultiLabelDataset
image_root: dataset/coco_ml/images
cls_label_path: dataset/coco_ml/train.txt
transform_ops:
- DecodeImage:
to_rgb: True
channel_first: False
- ResizeImage:
size: 448
interpolation: bilinear
backend: pil
- Cutout:
length: 224
fill_value: none
- RandAugmentV4:
- NormalizeImage:
scale: 1.0/255.0
mean: [0.485, 0.456, 0.406]
std: [0.229, 0.224, 0.225]
order: ''
sampler:
name: DistributedBatchSampler
batch_size: 64
drop_last: False
shuffle: True
loader:
num_workers: 8
use_shared_memory: True
Eval:
dataset:
name: MultiLabelDataset
image_root: dataset/coco_ml/images
cls_label_path: dataset/coco_ml/val.txt
transform_ops:
- DecodeImage:
to_rgb: True
channel_first: False
- ResizeImage:
size: 448
interpolation: bilinear
backend: pil
- NormalizeImage:
scale: 1.0/255.0
mean: [0.485, 0.456, 0.406]
std: [0.229, 0.224, 0.225]
order: ''
sampler:
name: DistributedBatchSampler
batch_size: 32
drop_last: False
shuffle: False
loader:
num_workers: 8
use_shared_memory: True
Infer:
infer_imgs: deploy/images/coco_000000570688.jpg
batch_size: 10
transforms:
- DecodeImage:
to_rgb: True
channel_first: False
- ResizeImage:
size: 448
interpolation: bilinear
backend: pil
- NormalizeImage:
scale: 1.0/255.0
mean: [0.485, 0.456, 0.406]
std: [0.229, 0.224, 0.225]
order: ''
- ToCHWImage:
PostProcess:
name: MultiLabelThreshOutput
threshold: 0.5
class_id_map_file: dataset/coco_ml/label.txt
Metric:
Train:
Eval:
- MultiLabelMAP:
# support list: integral, 11point
# default: integral
map_type: integral

168
ppcls/configs/MultiLabelCOCO/MLDecoder/ResNet50_ml_decoder_448.yaml 100644
View File

@ -0,0 +1,168 @@
# global configs
Global:
checkpoints: null
pretrained_model: null
output_dir: ./output/
device: gpu
save_interval: 10
eval_during_train: True
eval_interval: 1
epochs: 40
print_batch_step: 10
use_visualdl: False
# used for static mode and model export
image_shape: [3, 448, 448]
save_inference_dir: ./inference
# training model under @to_static
to_static: False
use_multilabel: True
# model ema
EMA:
decay: 0.9997
# mixed precision
AMP:
use_amp: True
use_fp16_test: False
scale_loss: 128.0
use_dynamic_loss_scaling: True
use_promote: False
# O1: mixed fp16, O2: pure fp16
level: O1
# model architecture
Arch:
name: ResNet50
class_num: 80
pretrained: True
use_ml_decoder: True
# ml-decoder head
MLDecoder:
query_num: 80 # default: 80, query_num <= class_num
in_channels: 2048
# loss function config for training/eval process
Loss:
Train:
- MultiLabelAsymmetricLoss:
weight: 1.0
gamma_pos: 0
gamma_neg: 4
clip: 0.05
disable_focal_loss_grad: True
Eval:
- MultiLabelAsymmetricLoss:
weight: 1.0
gamma_pos: 0
gamma_neg: 4
clip: 0.05
disable_focal_loss_grad: True
Optimizer:
name: AdamW
beta1: 0.9
beta2: 0.999
epsilon: 1e-8
weight_decay: 1e-4
one_dim_param_no_weight_decay: True
lr:
name: Cosine
learning_rate: 1e-4
eta_min: 1e-10
warmup_epoch: 5
warmup_start_lr: 1e-6
# data loader for train and eval
DataLoader:
Train:
dataset:
name: MultiLabelDataset
image_root: dataset/coco_ml/images
cls_label_path: dataset/coco_ml/train.txt
transform_ops:
- DecodeImage:
to_rgb: True
channel_first: False
- ResizeImage:
size: 448
interpolation: bilinear
backend: pil
- Cutout:
length: 224
fill_value: none
- RandAugmentV4:
- NormalizeImage:
scale: 1.0/255.0
mean: [0.485, 0.456, 0.406]
std: [0.229, 0.224, 0.225]
order: ''
sampler:
name: DistributedBatchSampler
batch_size: 64
drop_last: False
shuffle: True
loader:
num_workers: 8
use_shared_memory: True
Eval:
dataset:
name: MultiLabelDataset
image_root: dataset/coco_ml/images
cls_label_path: dataset/coco_ml/val.txt
transform_ops:
- DecodeImage:
to_rgb: True
channel_first: False
- ResizeImage:
size: 448
interpolation: bilinear
backend: pil
- NormalizeImage:
scale: 1.0/255.0
mean: [0.485, 0.456, 0.406]
std: [0.229, 0.224, 0.225]
order: ''
sampler:
name: DistributedBatchSampler
batch_size: 16
drop_last: False
shuffle: False
loader:
num_workers: 8
use_shared_memory: True
Infer:
infer_imgs: deploy/images/coco_000000570688.jpg
batch_size: 10
transforms:
- DecodeImage:
to_rgb: True
channel_first: False
- ResizeImage:
size: 448
interpolation: bilinear
backend: pil
- NormalizeImage:
scale: 1.0/255.0
mean: [0.485, 0.456, 0.406]
std: [0.229, 0.224, 0.225]
order: ''
- ToCHWImage:
PostProcess:
name: MultiLabelThreshOutput
threshold: 0.5
class_id_map_file: dataset/coco_ml/label.txt
Metric:
Train:
Eval:
- MultiLabelMAP:
# support list: integral, 11point
# default: integral
map_type: integral

2
ppcls/data/preprocess/ops/cutout.py
View File

@ -50,6 +50,6 @@ class Cutout(object):
random.randint(0, 255),
random.randint(0, 255)]
img = cv2.rectangle(img, (x1, y1), (x2, y2), fill_value, -1)
img = cv2.rectangle(np.array(img), (x1, y1), (x2, y2), fill_value, -1)
return img