mjq11302010044 / real-ce

Real-CE: A Benchmark for Chinese-English Scene Text Image Super-resolution (ICCV2023)

Jianqi Ma

real-ce's Issues

训练集的字体并未对齐

训练集的字体并未对齐，有的甚至相差很大，这是否有问题？

论文写的真不错，但git跑起来会有一些路径小问题，可以自己修复。但少了个net_g_100000.pth模型

代码中有一些是路径的问题。在basicsr/losses/losses.py 、 basicsr/metrics/recognition.py 等等文件中，有一些模型的加载路径写成了你自己的路径，比如这样：
/home/majianqi/workspace/BasicSR/basicsr/metrics/...

但实际上应该都用/basicsr/metrics/...就可以了。

另外再options下的yml里也有一些不存在的路径，需要自己处理一下。

上面的这些都好弄，但在options的yml文件中有大量的net_g_100000.pth模型引用。大佬能不能放出这个文件啊？

options of other models

Thank you for providing the options for the RRDB model. However, where can I find the configuration options for other models? Could you provide the configuration options for text super-resolution models and the ELAN model?
Additionally, how can I determine which pair of data to use for training on the RealCE dataset under the x2 condition? Is any code modification required?

sr_model.py里有个idx变量错误

Testing RealCE...
0image [00:00, ?image/s]
Traceback (most recent call last):
File "/root/autodl-tmp/Real-CE/basicsr/test.py", line 46, in
test_pipeline(root_path)
File "/root/autodl-tmp/Real-CE/basicsr/test.py", line 40, in test_pipeline
model.validation(test_loader, current_iter=opt['name'], tb_logger=None, save_img=opt['val']['save_img'])
File "/root/autodl-tmp/Real-CE/basicsr/models/base_model.py", line 48, in validation
self.nondist_validation(dataloader, current_iter, tb_logger, save_img)
File "/root/autodl-tmp/Real-CE/basicsr/models/sr_model.py", line 691, in nondist_validation
self.metric_results[metric] /= (idx + 1)
UnboundLocalError: local variable 'idx' referenced before assignment

weight of pretrained SR model

hello, will you public the pretrained RRDB SR weights, I notice that there are 3 .pth files in your google share link, however they seems not be the text Super Resolution weights.

Datasets always fail to download

。

cannot import name '_C' from 'basicsr.ops'

I‘ve run python setup.py develop, but the error `cannot import name '_C' from 'basicsr.ops' still existed. Would you tell me the way to resolve it?

The file `_C.cpython-310-x86_64-linux-gnu.so` cannot work with my environments.

After running the code python setup.py develop, I tried to run bash test_RRDB_RealCEx4.sh to test the model, but I get the errors information as following... I suppose the problem is that Real-CE/basicsr/ops/_C.cpython-310-x86_64-linux-gnu.so does not support my environment. However, I do not know what to do to recompile the corresponding .cpp file. Could you please open source the details about that. Thank you!

Traceback (most recent call last): File "/mnt/lustre/lihao5/Real-CE/basicsr/test.py", line 5, in <module> from basicsr.data import build_dataloader, build_dataset File "/mnt/lustre/lihao5/Real-CE/basicsr/__init__.py", line 3, in <module> from .archs import * File "/mnt/lustre/lihao5/Real-CE/basicsr/archs/__init__.py", line 16, in <module> _arch_modules = [importlib.import_module(f'basicsr.archs.{file_name}') for file_name in arch_filenames] File "/mnt/lustre/lihao5/Real-CE/basicsr/archs/__init__.py", line 16, in <listcomp> _arch_modules = [importlib.import_module(f'basicsr.archs.{file_name}') for file_name in arch_filenames] File "/mnt/cache/lihao5/miniconda3/envs/realce/lib/python3.10/importlib/__init__.py", line 126, in import_module return _bootstrap._gcd_import(name[level:], package, level) File "/mnt/lustre/lihao5/Real-CE/basicsr/archs/srresnet_arch.py", line 9, in <module> from basicsr.ops.roi_align_rotated import ROIAlignRotated File "/mnt/lustre/lihao5/Real-CE/basicsr/ops/roi_align_rotated.py", line 7, in <module> from basicsr.ops import _C ImportError: /mnt/lustre/lihao5/Real-CE/basicsr/ops/_C.cpython-310-x86_64-linux-gnu.so: undefined symbol: _ZN2at4_ops4view4callERKNS_6TensorEN3c108ArrayRefIlEE phoenix-srun: error: SH-IDC1-10-198-6-143: task 0: Exited with exit code 1

script about extracting annos

Hi,

Thanks for releasing this amazing TextSR benchmark.

I would like to apply this on other datasets. Could privide the script to extract annos information from images?

Looking forward to your reply~

How to inference?

How to inference?
I have some confuse, while using basicsr/models/sr_model.py and the model of 'RRDB_RealCE.pth',
the code:

print(img.shape)
output = self.net_g(img)
print(output.shape)

print:
torch.Size([1, 8120, 9792, 3])
torch.Size([1, 4, 39168, 12])

but the output shape is not a image, Is that right？Or any post-processing I need add?