自行训练的模型 转inference后 rec识别准确度低

[LiYifanyvonne]

#2470 问题下已经逐条排查过了，github上能搜到的#2080 #2782 #2873也都排查了
参数已对齐，检测还算准确，但是识别有很大问题
还有什么需要排查的点吗

举例来说：
python tools/infer/predict_system.py --image_dir="./try2/" --det_model_dir="./inference/db_mv3" --rec_model_dir="./inference/rec_ic" --use_angle_cls=false --rec_char_dict_path="./ppocr/utils/cchw_dict.txt" --rec_image_shape="3,32,100"
结果：

python tools/infer_rec.py -c configs/rec/rec_icdar15_train.yml -o Global.pretrained_model=./output/rec/ic15/best_accuracy Global.load_static_weights=false Global.infer_img=./try2/
结果：
./try2/20210926174943.pngsub_pic0_1.png 剩余期限 0.83949465

@LDOUBLEV

[LiYifanyvonne]

rec的yml参数
Global:
use_gpu: false
epoch_num: 1200
log_smooth_window: 20
print_batch_step: 20
save_model_dir: ./output/rec/ic15/
save_epoch_step: 360
#evaluation is run every 2000 iterations
eval_batch_step: [0, 360]
cal_metric_during_train: True
pretrained_model: ./pretrain_models/rec_mv3_none_bilstm_ctc_v2.0_train/best_accuracy
checkpoints:
save_inference_dir:
use_visualdl: False
infer_img: doc/imgs_words_en/word_10.png
#for data or label process
character_dict_path: ./ppocr/utils/cchw_dict.txt
character_type: ch
max_text_length: 25
infer_mode: False
use_space_char: True
save_res_path: ./output/rec/predicts_ic15.txt

Optimizer:
name: Adam
beta1: 0.9
beta2: 0.999
lr:
learning_rate: 0.0005
regularizer:
name: 'L2'
factor: 0

Architecture:
model_type: rec
algorithm: CRNN
Transform:
Backbone:
name: MobileNetV3
scale: 0.5
model_name: large
Neck:
name: SequenceEncoder
encoder_type: rnn
hidden_size: 96
Head:
name: CTCHead
fc_decay: 0

Loss:
name: CTCLoss

PostProcess:
name: CTCLabelDecode

Metric:
name: RecMetric
main_indicator: acc

Train:
dataset:
name: SimpleDataSet
data_dir: ./train_data/rec/
label_file_list: ["./train_data/rec/rec_gt_train.txt"]
transforms:
- DecodeImage: # load image
img_mode: BGR
channel_first: False
- CTCLabelEncode: # Class handling label
- RecResizeImg:
image_shape: [3, 32, 100]
- KeepKeys:
keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
loader:
shuffle: True
batch_size_per_card: 2
drop_last: True
num_workers: 1
use_shared_memory: False

Eval:
dataset:
name: SimpleDataSet
data_dir: ./train_data/rec
label_file_list: ["./train_data/rec/rec_gt_test.txt"]
transforms:
- DecodeImage: # load image
img_mode: BGR
channel_first: False
- CTCLabelEncode: # Class handling label
- RecResizeImg:
image_shape: [3, 32, 100]
- KeepKeys:
keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
loader:
shuffle: False
drop_last: False
batch_size_per_card: 8
num_workers: 1
use_shared_memory: False

[LiYifanyvonne]

paddleOCR 代码内对齐的参数
else:
return argparse.Namespace(
use_gpu=False,
ir_optim=True,
use_tensorrt=False,
gpu_mem=8000,
image_dir='',
det_algorithm='DB',
det_model_dir="./inference/db_mv3",#None,
det_limit_side_len=960,
det_limit_type='max',
det_db_thresh=0.3,
det_db_box_thresh=0.6,
det_db_unclip_ratio=1.5,
use_dilation=False,
det_db_score_mode="fast",
det_east_score_thresh=0.8,
det_east_cover_thresh=0.1,
det_east_nms_thresh=0.2,
rec_algorithm='CRNN',
rec_model_dir="./inference/rec_ic",#None,
rec_image_shape="3, 32, 100",#
rec_char_type='ch',
rec_batch_num=2,
max_text_length=25,
rec_char_dict_path="./ppocr/utils/cchw_dict.txt",#None,
use_space_char=True,
drop_score=0.5,
cls_model_dir=None,
cls_image_shape="3, 48, 192",
label_list=['0', '180'],
cls_batch_num=6,
cls_thresh=0.9,
enable_mkldnn=False,
use_zero_copy_run=False,
use_pdserving=False,
lang='ch',
det=True,
rec=True,
use_angle_cls=False)

[ioracion]

你用predict_system是检测+识别，而infer_rec是纯识别。所以可能是加了检测模型才导致结果不对。用predict_rec跟infer_rec对比试试呢

[LiYifanyvonne]

你用predict_system是检测+识别，而infer_rec是纯识别。所以可能是加了检测模型才导致结果不对。用predict_rec跟infer_rec对比试试呢

用语句python tools/infer/predict_rec.py --image_dir="./try3/" --rec_model_dir="./inference/rec_ic" --rec_char_dict_path="./ppocr/utils/cchw_dict.txt" --rec_image_shape="3,32,100"

结果：

的确可能是检测的问题，但是用predict_system时，检测框选的是正确的。
而且检测参数也对齐过的

Global:
use_gpu: false
epoch_num: 30
log_smooth_window: 20
print_batch_step: 10
save_model_dir: ./output/db_mv3/
save_epoch_step: 15
#evaluation is run every 2000 iterations
eval_batch_step: [0,15]
cal_metric_during_train: False
pretrained_model: ./pretrain_models/ch_ppocr_mobile_v2.0_det_train/best_accuracy
checkpoints:
save_inference_dir:
use_visualdl: False
infer_img: doc/imgs_en/img_10.jpg
save_res_path: ./output/det_db/predicts_db.txt

Architecture:
model_type: det
algorithm: DB
Transform:
Backbone:
name: MobileNetV3
scale: 0.5
model_name: large
Neck:
name: DBFPN
out_channels: 256
Head:
name: DBHead
k: 50

Loss:
name: DBLoss
balance_loss: true
main_loss_type: DiceLoss
alpha: 5
beta: 10
ohem_ratio: 3

Optimizer:
name: Adam
beta1: 0.9
beta2: 0.999
lr:
learning_rate: 0.001
regularizer:
name: 'L2'
factor: 0

PostProcess:
name: DBPostProcess
thresh: 0.3
box_thresh: 0.6
max_candidates: 1000
unclip_ratio: 1.5

Metric:
name: DetMetric
main_indicator: hmean

Train:
dataset:
name: SimpleDataSet
data_dir: ./train_data/
label_file_list:
- ./train_data/train_label.txt
ratio_list: [1.0]
transforms:
- DecodeImage: # load image
img_mode: BGR
channel_first: False
- DetLabelEncode: # Class handling label
- IaaAugment:
augmenter_args:
- { 'type': Fliplr, 'args': { 'p': 0.5 } }
- { 'type': Affine, 'args': { 'rotate': [-10, 10] } }
- { 'type': Resize, 'args': { 'size': [0.5, 3] } }
- EastRandomCropData:
size: [640, 640]
max_tries: 50
keep_ratio: true
- MakeBorderMap:
shrink_ratio: 0.4
thresh_min: 0.3
thresh_max: 0.7
- MakeShrinkMap:
shrink_ratio: 0.4
min_text_size: 8
- NormalizeImage:
scale: 1./255.
mean: [0.485, 0.456, 0.406]
std: [0.229, 0.224, 0.225]
order: 'hwc'
- ToCHWImage:
- KeepKeys:
keep_keys: ['image', 'threshold_map', 'threshold_mask', 'shrink_map', 'shrink_mask'] # the order of the dataloader list
loader:
shuffle: True
drop_last: False
batch_size_per_card: 16
num_workers: 1
use_shared_memory: False

Eval:
dataset:
name: SimpleDataSet
data_dir: ./train_data/
label_file_list:
- ./train_data/test_label.txt
transforms:
- DecodeImage: # load image
img_mode: BGR
channel_first: False
- DetLabelEncode: # Class handling label
# - DetResizeForTest:
#image_shape: [736, 1280]
- NormalizeImage:
scale: 1./255.
mean: [0.485, 0.456, 0.406]
std: [0.229, 0.224, 0.225]
order: 'hwc'
- ToCHWImage:
- KeepKeys:
keep_keys: ['image', 'shape', 'polys', 'ignore_tags']
loader:
shuffle: False
drop_last: False
batch_size_per_card: 1 # must be 1
num_workers: 1
use_shared_memory: False

[ioracion]

emmm这样的图片其实没有必要用检测模型，字基本占了整张图，去掉检测模型速度还更快

[paddle-bot-old]

Since you haven't replied for more than 3 months, we have closed this issue/pr.
If the problem is not solved or there is a follow-up one, please reopen it at any time and we will continue to follow up.
It is recommended to pull and try the latest code first.
由于您超过三个月未回复，我们将关闭这个issue/pr。
若问题未解决或有后续问题，请随时重新打开（建议先拉取最新代码进行尝试），我们会继续跟进。