How can I resume training when I have previous weights.

How should I proceed to train on my own data.
I didn't understand the meaning of separate gt text file for each image. What should the image be and the corresponding gt text file content?
Thanks a lot!

Get this error on running python3 eval.py

make: Entering directory '/home/sam/CVPlayground/FOTS/utils/lanms'
make: 'adaptor.so' is up to date.
make: Leaving directory '/home/sam/CVPlayground/FOTS/utils/lanms'
Loading checkpoint: models/retrained_model.pth.tar ...
Traceback (most recent call last):
  File "eval.py", line 54, in <module>
    main(args)
  File "eval.py", line 33, in main
    model = load_model(model_path, with_gpu)
  File "eval.py", line 15, in load_model
    checkpoints = torch.load(model_path)
  File "/home/sam/.local/lib/python3.6/site-packages/torch/serialization.py", line 368, in load
    return _load(f, map_location, pickle_module)
  File "/home/sam/.local/lib/python3.6/site-packages/torch/serialization.py", line 542, in _load
    result = unpickler.load()
ModuleNotFoundError: No module named 'logger.logger'

How to solve this?

Have you tried Thz architecture on synth90K or Synth800K?

hey, i've tried to train your model with RCTW dataset.
It works.
However:

Epoch 1 / 30000
126
poly in wrong direction
poly in wrong direction
poly in wrong direction
poly in wrong direction
poly in wrong direction

Then I had to stop.

The label looks as:

390,902,1856,902,1856,1225,390,1225,0,"金氏眼镜"
1875,1170,2149,1170,2149,1245,1875,1245,0,"创于1989"
2054,1277,2190,1277,2190,1323,2054,1323,0,"城建店"
...

Seems that it works in other models' training like ctpr.
Is there anything wrong? I'm really confused.

Hi @xieyufei1993, when training model I got an error such as:
(fots) loitg@loitg-Precision-T3600:~/Desktop/tu_workspace/FOTS$ python main_train.py
use config:
initial_epoch 0
epoch_num 30000
lr 0.001
decay 0.0005
use_gpu True
batch_size 64
num_workers 10
optmizer RMSprop
betas (0.5, 0.999)
epsilon 0.0001
shrink_side_ratio 0.6
shrink_ratio 0.2
model FOTS
patience 2
load_weights False
lambda_inside_score_loss 4.0
lambda_side_vertex_code_loss 1.0
lambda_side_vertex_coord_loss 1.0
load_model_path checkpoints/model.pth
save_path save_model/
total_img 16243
validation_split_ratio 0.1
max_train_img_size 736
max_predict_img_size 2400
parse <bound method parse of <config.DefaultConfig object at 0x7fe4ff6d5cc0>>
end the parse!!!

Epoch 1 / 30000
Traceback (most recent call last):
File "main_train.py", line 95, in
main()
File "main_train.py", line 90, in main
save_step=5, weight_decay=weight_decay)
File "main_train.py", line 32, in train
for i, (img, score_map, geo_map, training_mask) in enumerate(trainloader):
File "/home/loitg/miniconda3/envs/fots/lib/python3.6/site-packages/torch/utils/data/dataloader.py", line 582, in next
return self._process_next_batch(batch)
File "/home/loitg/miniconda3/envs/fots/lib/python3.6/site-packages/torch/utils/data/dataloader.py", line 608, in _process_next_batch
raise batch.exc_type(batch.exc_msg)
RuntimeError: Traceback (most recent call last):
File "/home/loitg/miniconda3/envs/fots/lib/python3.6/site-packages/torch/utils/data/_utils/worker.py", line 99, in _worker_loop
samples = collate_fn([dataset[i] for i in batch_indices])
File "/home/loitg/Desktop/tu_workspace/FOTS/data/dataset.py", line 672, in collate_fn
images = torch.stack(images, 0)
RuntimeError: expected a non-empty list of Tensors

How to fix this error ?. I hope to see your reply soon.

Hi，我是FOTS的作者，由于公司限制和实现平台我们难以公开FOTS代码。看到了你在gitlab复现的FOTS，请问现在结果如何？有什么困难吗？希望可以帮助你一起复现出FOTS的精度

Hi ,Thanks a lot for sharing Your work.

Could you please upload the model.

@xieyufei1993 No module named 'pretrainedmodels'

@xieyufei1993
What results is it possible to achieve with the code on ICDAR 2015/ICDAR 2017?
@MaxwellRebo, @Samleo8, @thuyngch, may be you can comment, since you did some updates on your forks. What results did you manage to get?

Traceback (most recent call last):
File "eval.py", line 6, in
from models.FOTS import FOTS
File "/home/kobe/workspace/FOTS/models/init.py", line 1, in
from models.FOTS import FOTS
File "/home/kobe/workspace/FOTS/models/FOTS.py", line 3, in
import pretrainedmodels as pm
ModuleNotFoundError: No module named 'pretrainedmodels'
I don't know how to fix the above problem.

I did not find it.

Could share the trained models on the ICDAR?

Thank you very much

Hi,
I cannot find the link for the trained model file. Could you add it on the page?

thx a lot

No recognized branch was found

@xieyufei1993 sorry to border you,but i wonder why it always remind me that poly in wrong direction when training

please add a license to the repo so it can be used by others without legal doubts. Even if you are stuck with RoIRotate someone else might be able to complete the work if you add a proper license. MIT or Apache 2 seem to be the most popular choices.