Custom image Inference

[kdh-awraw1019]

I want to inference about a custom image.

this image is not in 'Prepare datasets' ex) ic17, ic15, ...

How can I inference a custom image?

How should I change it here? ▼▼

./config/pan_pp/pan_pp_r18_ic15_joint_train.py

data = dict(
batch_size=16,
train=dict(
type='PAN_PP_JointTrain',
split='train',
is_transform=True,
img_size=736,
short_size=736,
kernel_scale=0.5,
read_type='pil',
with_rec=True
),
test=dict(
type='PAN_PP_IC15',
split='test',
short_size=720,
read_type='pil',
with_rec=True
)
)

[EMRAN-SALEH-CORSEARCH]

Yeah, Same here. I want ro do inference of PAN++ on custom image. I do not see that in the repo.

[kdh-awraw1019]

Yeah, Same here. I want ro do inference of PAN++ on custom image. I do not see that in the repo.

I did modified some codes and inference with my cunstom data.

1. pan_pp.pytorch/dataset/pan_pp/pan_pp_joint_train.py

import math
import random
import string

import cv2
import mmcv
import numpy as np
import Polygon as plg
import pyclipper
import scipy.io as scio
import torch
import torchvision.transforms as transforms
from PIL import Image
from torch.utils import data

from .coco_text import COCO_Text

EPS = 1e-6
synth_root_dir = './data/SynthText/'
synth_train_data_dir = synth_root_dir
synth_train_gt_path = synth_root_dir + 'gt.mat'

ic17_root_dir = './data/ICDAR2017MLT/'
ic17_train_data_dir = ic17_root_dir + 'ch8_training_images/'
ic17_train_gt_dir = ic17_root_dir +
'ch8_training_localization_transcription_gt_v2/'

ct_root_dir = './data/COCO-Text/'
ct_train_data_dir = ct_root_dir + 'train2014/'
ct_train_gt_path = ct_root_dir + 'COCO_Text.json'

'''
ic15_root_dir = './data/ICDAR2015/Challenge4/'
ic15_train_data_dir = ic15_root_dir + 'ch4_training_images/'
ic15_train_gt_dir = ic15_root_dir +
'ch4_training_localization_transcription_gt/'
'''
############ modified #######
ic15_root_dir = your dataset path location ##
ic15_test_data_dir = ic15_root_dir + 'imgs/'
ic15_test_gt_dir = ic15_root_dir + 'gts/'

##################################

2. pan_pp.pytorch/configs/pan_pp/pan_pp_r18_ic15_joint_train.py
   data = dict(
   batch_size=1,
   train=dict(
   type='PAN_PP_JointTrain',
   split='train',
   is_transform=True,
   img_size=736,# 736,
   short_size=736,# 736,
   kernel_scale=0.5,
   read_type='pil',
   with_rec=True
   ),
   test=dict(
   type='PAN_PP_IC15', ## my custom dataset's format is ic15
   split='test',
   short_size=96,# default : 720,
   read_type='pil',
   with_rec=True
   )
   )

[Devin521314]

How to predict a new image using the training weight?it doesn't work below.

python test.py config/pan/pan_r18_ic15.py checkpoints/pan_r18_ic15/checkpoint.pth.tar
cd eval/
./eval_ic15.sh

please inform me with qushanghui@niii.com or wechat SanQian-2012,thanks you so much.