# ABCNetv1 & ABCNetv2
[ABCNetv1](https://openaccess.thecvf.com/content_CVPR_2020/html/Liu_ABCNet_Real-Time_Scene_Text_Spotting_With_Adaptive_Bezier-Curve_Network_CVPR_2020_paper.html) is an efficient end-to-end scene text spotting framework over 10x faster than previous state of the art. It's published in IEEE Conf. Comp Vis Pattern Recogn.'2020 as an oral paper. [ABCNetv2](https://ieeexplore.ieee.org/document/9525302) is published in TPAMI.
## Models
### Experimental resutls on CTW1500:
Name | inf. time | e2e-None-hmean | e2e-Full-hmean | det-hmean | download
--- |:---:|:---:|:---:|:---:|:---:
[v1-CTW1500-finetune](CTW1500/attn_R_50.yaml) | 8.7 FPS | 53.2 | 74.7 | 84.4 | [pretrained-model](https://universityofadelaide.box.com/shared/static/okeo5pvul5v5rxqh4yg8pcf805tzj2no.pth)
[v2-CTW1500-finetune](CTW1500/v2_attn_R_50.yaml) | 7.2 FPS | 57.7 | 75.8 | 85.0 | [finetuned-model](https://drive.google.com/file/d/12HV1dHjw1POdhOiHXPPXcGnjyp-3IuQv/view?usp=sharing)
### Experimental resutls on TotalText:
Config | inf. time | e2e-None-hmean | e2e-Full-hmean | det-hmean | download
--- |:---------:|:---------:|:---------:|:---------:|:---:
[v1-pretrain](Pretrain/attn_R_50.yaml) | 11.3 FPS | 58.1 | 75.5 | 80.0 | [pretrained-model](https://cloudstor.aarnet.edu.au/plus/s/dEzxhTlEumICiq0/download)
[v1-totaltext-finetune](TotalText/attn_R_50.yaml) | 11.3 FPS | 67.1 | 81.1 | 86.0 | [finetuned-model](https://cloudstor.aarnet.edu.au/plus/s/tYsnegjTs13MwwK/download)
[v2-pretrain](Pretrain/v2_attn_R_50.yaml) | 7.8 FPS | 63.5 | 78.4 | 83.7 | [pretrained-model](https://drive.google.com/file/d/1v5C9klxBuNVBaLVxZRCy1MYnwEu0F25q/view?usp=sharing)
[v2-totaltext-finetune](TotalText/v2_attn_R_50.yaml) | 7.7 FPS | 71.8 | 83.4 | 87.2 | [finetuned-model](https://drive.google.com/file/d/1jR5-A-7ITvjdSx3kWVE9bMgh_biMsqcR/view?usp=sharing)
### Experimental resutls on [ICDAR2015](https://rrc.cvc.uab.es/?ch=4):
Name | e2e-None | e2e-Generic | e2e-Weak | e2e-Strong | det-hmean | download
--- |:---:|:---:|:---:|:---:|:---:|:---:
[v1-icdar2015-pretrain](Pretrain/v1_ic15_attn_R_50.yaml) | 38.0 | 50.8 | 59.0 | 65.8 | 83.2 | [pretrained-model](https://drive.google.com/file/d/1MZab_ftY8qGCurW1rwZBx5ftquZgcf4e/view?usp=sharing)
[v1-icdar2015-finetune](ICDAR2015/v1_attn_R_50.yaml) | 57.1 | 66.8 | 74.1 | 79.2 | 86.8 | [finetuned-model](https://drive.google.com/file/d/15eEctI4CqTxtcMAMcYiHIysYw3l53BGQ/view?usp=sharing)
[v2-icdar2015-pretrain](Pretrain/v2_ic15_attn_R_50.yaml) | 59.5 | 69.0 | 75.8 | 80.8 | 86.2 | [pretrained-model](https://drive.google.com/file/d/17xIB064Jlq31z875POrw9a3aDmg04C3y/view?usp=sharing)
[v2-icdar2015-finetune](ICDAR2015/v2_attn_R_50.yaml) | 66.3 | 73.2 | 78.8 | 83.7 | 88.2 | [finetuned-model](https://drive.google.com/file/d/1bxVxu7kX13S1_xYvCfUfomO8hSZGNZUl/view?usp=sharing)
### Experimental resutls on [ReCTS](https://rrc.cvc.uab.es/?ch=12):
Name | inf. time | det-recall | det-precision | det-hmean | 1 - NED | download
--- |:---:|:---:|:---:|:---:|:---:|:---:
[v2-Chinese-pretrained](Pretrain/v2_chn_attn_R_50.yaml) | -| - | - | - | - | [pretrained-model](https://drive.google.com/file/d/1AU8yAMNm2H8ryB7uIvp2HUHpCso7eyNH/view?usp=sharing)
[v2-ReCTS-finetune](ReCTS/v2_chn_attn_R_50.yaml) | 8 FPS | 87.9 | 92.9 | 90.33 | 63.9 | [finetuned-model](https://drive.google.com/file/d/1YTlC5jkh6y3g1RRc_hDs4m_tcU2J20fe/view?usp=sharing)
### Experimental resutls on [MSRA-TD500](http://www.iapr-tc11.org/mediawiki/index.php/MSRA_Text_Detection_500_Database_%28MSRA-TD500%29):
Name | det-recall | det-precision | det-hmean | download
--- |:---:|:---:|:---:|:---:
[v2-TD500-finetune](https://github.com/aim-uofa/AdelaiDet/issues/537) | 81.9 | 89.0 | 85.3 | [finetuned-model](https://github.com/aim-uofa/AdelaiDet/issues/537)
* Note the pretrained model for TD500 is the Chinese pretrained used for ReCTS. As MSRA-TD is a det. only dataset, a small amount of [modification](https://github.com/aim-uofa/AdelaiDet/issues/537) is needed.
## Quick Start (ABCNetv1)
### Inference with our trained Models
1. Select the model and config file above, for example, `configs/BAText/CTW1500/attn_R_50.yaml`.
2. Run the demo with
```
wget -O ctw1500_attn_R_50.pth https://universityofadelaide.box.com/shared/static/okeo5pvul5v5rxqh4yg8pcf805tzj2no.pth
python demo/demo.py \
--config-file configs/BAText/CTW1500/attn_R_50.yaml \
--input datasets/CTW1500/ctwtest_text_image/ \
--opts MODEL.WEIGHTS ctw1500_attn_R_50.pth
```
or
```
wget -O tt_attn_R_50.pth https://cloudstor.aarnet.edu.au/plus/s/tYsnegjTs13MwwK/download
python demo/demo.py \
--config-file configs/BAText/TotalText/attn_R_50.yaml \
--input datasets/totaltext/test_images/ \
--opts MODEL.WEIGHTS tt_attn_R_50.pth
```
or
```
# Download v1_ic15_finetuned.pth above
python demo/demo.py \
--config-file configs/BAText/ICDAR2015/v1_attn_R_50.yaml \
--input datasets/icdar2015/test_images \
--opts MODEL.WEIGHTS v1_ic15_finetuned.pth
```
### Train Your Own Models
To train a model with "train_net.py", first setup the corresponding datasets following
[datasets/README.md](../../datasets/README.md) or using the following script:
```
cd datasets/
wget https://drive.google.com/file/d/1we4iwZNA80q-yRoEKqB66SuTa1tPbhZu/view?usp=sharing -O totaltext.zip
unzip totaltext.zip
rm totaltext.zip
wget https://drive.google.com/file/d/1ntlnlnQHZisDoS_bgDvrcrYFomw9iTZ0/view?usp=sharing -O CTW1500.zip
unzip CTW1500.zip
rm CTW1500.zip
wget https://drive.google.com/file/d/1J94245rU-s7KTecNQRD3KXG04ICZhL9z/view?usp=sharing -O icdar2015.zip
unzip icdar2015.zip
rm icdar2015.zip
mkdir evaluation
cd evaluation
wget -O gt_ctw1500.zip https://cloudstor.aarnet.edu.au/plus/s/xU3yeM3GnidiSTr/download
wget -O gt_totaltext.zip https://cloudstor.aarnet.edu.au/plus/s/SFHvin8BLUM4cNd/download
wget -O gt_icdar2015.zip https://drive.google.com/file/d/1wrq_-qIyb_8dhYVlDzLZTTajQzbic82Z/view?usp=sharing
```
* Note (synthetic and mlt2017 datasets need to be downloaded through [datasets/README.md](../../datasets/README.md).)
You can also prepare your custom dataset following the [example scripts](https://universityofadelaide.box.com/s/phqfzpvhe0obmkvn17akn9qw47u1m44i).
Pretrainining with synthetic data (For Totaltext and CTW1500):
```
OMP_NUM_THREADS=1 python tools/train_net.py \
--config-file configs/BAText/Pretrain/attn_R_50.yaml \
--num-gpus 4 \
OUTPUT_DIR text_pretraining/attn_R_50
```
Pretrainining with synthetic data (For ICDAR2015):
```
OMP_NUM_THREADS=1 python tools/train_net.py \
--config-file configs/BAText/Pretrain/v1_ic15_attn_R_50.yaml \
--num-gpus 4 \
OUTPUT_DIR text_pretraining/v1_ic15_attn_R_50
```
Finetuning on TotalText:
```
OMP_NUM_THREADS=1 python tools/train_net.py \
--config-file configs/BAText/TotalText/attn_R_50.yaml \
--num-gpus 4 \
MODEL.WEIGHTS text_pretraining/attn_R_50/model_final.pth
```
Finetuning on CTW1500:
```
OMP_NUM_THREADS=1 python tools/train_net.py \
--config-file configs/BAText/CTW1500/attn_R_50.yaml \
--num-gpus 4 \
MODEL.WEIGHTS text_pretraining/attn_R_50/model_final.pth
```
Finetuning on ICDAR2015:
```
OMP_NUM_THREADS=1 python tools/train_net.py \
--config-file configs/BAText/ICDAR2015/v1_attn_R_50.yaml \
--num-gpus 4 \
MODEL.WEIGHTS text_pretraining/v1_ic15_attn_R_50/model_final.pth
```
### Evaluate on Trained Model
Download test GT [here](../../datasets/README.md) so that the directory has the following structure:
```
datasets
|_ evaluation
| |_ gt_totaltext.zip
| |_ gt_ctw1500.zip
| |_ gt_icdar2015.zip
```
Producing both (w/wo lexion) e2e and detection results on CTW1500:
```
wget -O ctw1500_attn_R_50.pth https://universityofadelaide.box.com/shared/static/okeo5pvul5v5rxqh4yg8pcf805tzj2no.pth
python tools/train_net.py \
--config-file configs/BAText/CTW1500/attn_R_50.yaml \
--eval-only \
MODEL.WEIGHTS ctw1500_attn_R_50.pth
```
or Totaltext:
```
wget -O tt_attn_R_50.pth https://cloudstor.aarnet.edu.au/plus/s/tYsnegjTs13
没有合适的资源?快使用搜索试试~ 我知道了~
SOLOV2代码和预训练模型
共1078个文件
py:527个
yaml:284个
md:83个
需积分: 0 7 下载量 77 浏览量
2023-04-18
22:56:08
上传
评论 1
收藏 211.26MB RAR 举报
温馨提示
SOLOV2源代码和相关预训练模型
资源推荐
资源详情
资源评论
收起资源包目录
SOLOV2代码和预训练模型 (1078个子文件)
pkg_helpers.bash 2KB
setup.cfg 931B
.clang-format 2KB
cocoeval.cpp 20KB
BezierAlign_cpu.cpp 17KB
ROIAlignRotated_cpu.cpp 16KB
torchscript_traced_mask_rcnn.cpp 4KB
caffe2_mask_rcnn.cpp 4KB
vision.cpp 3KB
nms_rotated_cpu.cpp 2KB
vision.cpp 2KB
box_iou_rotated_cpu.cpp 1KB
vision.cpp 429B
custom.css 511B
deform_conv_cuda_kernel.cu 43KB
deform_conv_cuda.cu 31KB
DefROIAlign_cuda.cu 15KB
BezierAlign_cuda.cu 15KB
ROIAlignRotated_cuda.cu 14KB
SwapAlign2Nat_cuda.cu 13KB
ml_nms.cu 5KB
nms_rotated_cuda.cu 5KB
box_iou_rotated_cuda.cu 4KB
cuda_version.cu 622B
cuda_version.cu 122B
Dockerfile 2KB
deploy.Dockerfile 2KB
.flake8 459B
.gitignore 636B
.gitignore 556B
.gitignore 7B
.gitignore 6B
.gitignore 5B
box_iou_rotated_utils.h 11KB
deform_conv.h 8KB
cocoeval.h 3KB
BezierAlign.h 3KB
ROIAlignRotated.h 3KB
DefROIAlign.h 3KB
SwapAlign2Nat.h 1KB
nms_rotated.h 1KB
box_iou_rotated.h 988B
ml_nms.h 787B
lazyconfig.jpg 64KB
levenshtein.js 2KB
instances_train2014.json 11.25MB
instances_val2014.json 1.95MB
LICENSE 10KB
LICENSE 1KB
Makefile 634B
Makefile 630B
MODEL_ZOO.md 57KB
DENSEPOSE_IUV.md 30KB
DENSEPOSE_DATASETS.md 19KB
DENSEPOSE_CSE.md 16KB
datasets.md 14KB
README.md 14KB
README_ori.md 12KB
INSTALL.md 12KB
TOOL_APPLY_NET.md 10KB
models.md 9KB
README.md 9KB
augmentation.md 8KB
deployment.md 8KB
benchmarks.md 7KB
README.md 7KB
lazyconfigs.md 7KB
BOOTSTRAPPING_PIPELINE.md 7KB
README.md 6KB
extend.md 6KB
README.md 6KB
README.md 5KB
README.md 5KB
data_loading.md 5KB
README.md 5KB
README.md 4KB
compatibility.md 4KB
README.md 4KB
TOOL_QUERY_DB.md 4KB
write-models.md 4KB
README.md 4KB
README.md 4KB
README.md 4KB
README.md 4KB
CONTRIBUTING.md 4KB
README.md 3KB
MODEL_ZOO.md 3KB
training.md 3KB
README.md 3KB
README.md 3KB
README.md 3KB
GETTING_STARTED.md 3KB
GETTING_STARTED.md 3KB
RELEASE_2021_03.md 3KB
README.md 3KB
README.md 3KB
evaluation.md 3KB
changelog.md 3KB
configs.md 3KB
README.md 2KB
共 1078 条
- 1
- 2
- 3
- 4
- 5
- 6
- 11
资源评论
德卡先生的信箱
- 粉丝: 24
- 资源: 16
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功