jamycheung / matchformer Goto Github PK

View Code? Open in Web Editor NEW

167.0 167.0 10.0 199 KB

Repository of MatchFormer

License: Apache License 2.0

Python 100.00%

matchformer's People

Contributors

Stargazers

Watchers

Forkers

felipecadar cys1102 chqwer2 gabbysuwichaya dl-vit suke0 presunshine chjxu mooncake199809 sirkrisp

matchformer's Issues

Demo on an image pair request

Hi,

could you provide a small demo to just run the feature matching on a single image pair and provide the correspondences?

Thanks in advance!

感谢您杰出的工作！我想请问您论文中涉及到ClusterGNN: Cluster-based Coarse-to-Fine Graph Neural Network for Efficient
Feature Matching（cvpr2022）部分的实验数据是怎么获得的，据我所知，原作者并没有开源代码，您自己复现了它么？能否和我分享一下您复现的这部分工作？或者您知道哪里有相关的开源，万分感谢！我的邮箱是[email protected]或者[email protected]

Dataset Structure

Hi~ Thx for your excellent work!

I now have some problems when running the test code.It may be caused by the wrong file structure.

In fact, I'm not familiar with the structure of the two data sets of scannet and megadepth. I don't know where to download the megadepth_test_1500.txt file

MatchFormer/config/data/megadepth_test_1500.py

Line 8 in 5f40637

cfg.DATASET.TEST_LIST_PATH = f"{TEST_BASE_PATH}/megadepth_test_1500.txt"

Could you give me some help on downloading and organizing the dataset structure? Or modify the readme file to make the data format configuration easier?

Looking forward to your reply, and wish you have a nice day!

Fail to run model

I want to use the matcher outside the test script, but i'm unable to build the model :/

the function Matchformer_SEA_lite() asks for a config in the first argument (line)

but the build_backbone function, does not pass any config ahead (line)

Code

config = get_cfg_defaults()

# adjust lite SEA model config:
config.MATCHFORMER.BACKBONE_TYPE = 'litesea'
config.MATCHFORMER.SCENS = 'outdoor'
config.MATCHFORMER.RESOLUTION = (8,4)

model = PL_LoFTR(config, pretrained_ckpt="./pretrained/indoor-lite-LA.ckpt", dump_dir="/tmp")

Traceback

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
/tmp/ipykernel_38/1928295852.py in <module>
      6 config.MATCHFORMER.RESOLUTION = (8,4)
      7 
----> 8 model = PL_LoFTR(config, pretrained_ckpt="./pretrained/indoor-lite-LA.ckpt", dump_dir="/tmp")

/kaggle/working/MatchFormer/model/lightning_loftr.py in __init__(self, config, pretrained_ckpt, profiler, dump_dir)
     27 
     28         # Matcher: LoFTR
---> 29         self.matcher = Matchformer(config=_config['matchformer'])
     30 
     31         # Pretrained weights

/kaggle/working/MatchFormer/model/matchformer.py in __init__(self, config)
     14         # Misc
     15         self.config = config
---> 16         self.backbone = build_backbone(config)
     17         self.coarse_matching = CoarseMatching(config['match_coarse'])
     18         self.fine_preprocess = FinePreprocess(config)

/kaggle/working/MatchFormer/model/backbone/__init__.py in build_backbone(config)
     11         return Matchformer_LA_large()
     12     elif config['backbone_type'] == 'litesea':
---> 13         return Matchformer_SEA_lite()
     14     elif config['backbone_type'] == 'largesea':
     15         return Matchformer_SEA_large()

TypeError: __init__() missing 1 required positional argument: 'config'

Am I doing something terrible wrong ?

Missing dataset data and index files

The dataset resources provided by Loftr do not seem to include scan/index/scan net_ How can I obtain the test. txt and scan/inedx/test. npz files?

Using Matchformer with my own Images

Hi I'm looking to use Matchformer and I can't quite figure out how to test the model on my own images, could you include some instructions on how to do so?

the scannet_test_1500.py test path is wrong

感谢你的工作，这个test demo根本行不通。以scannet为例，scannet_test.txt和test.npz如何的来，原来的LoFTR根本没有这些，你自己测试过了吗？

where is training code?

Hi, thank you for your great job. But where is training code? why not open?

RuntimeError: Default process group has not been initialized, please make sure to call init_process_group.

Hi! While trying to run the test in a jupyter notebook, I was able to initialise the model with

pl.seed_everything(config.TRAINER.SEED)  # reproducibility
model = PL_LoFTR(config, pretrained_ckpt="../input/matchformer/MatchFormer/model/weights/outdoor-lite-SEA.ckpt", dump_dir="dump/")
# lightning data
data_module = MultiSceneDataModule(args, config)
# lightning trainer
trainer = pl.Trainer.from_argparse_args(args, logger=False, accelerator="gpu", gpus=1)

However, upon running
trainer.test(model, datamodule=data_module, verbose=True)

I am getting


RuntimeError                              Traceback (most recent call last)
/tmp/ipykernel_33/3924240025.py in <module>
----> 1 trainer.test(model, datamodule=data_module, verbose=True)

/opt/conda/lib/python3.7/site-packages/pytorch_lightning/trainer/trainer.py in test(self, model, dataloaders, ckpt_path, verbose, datamodule)
    934         """
    935         self.strategy.model = model or self.lightning_module
--> 936         return self._call_and_handle_interrupt(self._test_impl, model, dataloaders, ckpt_path, verbose, datamodule)
    937 
    938     def _test_impl(

/opt/conda/lib/python3.7/site-packages/pytorch_lightning/trainer/trainer.py in _call_and_handle_interrupt(self, trainer_fn, *args, **kwargs)
    719                 return self.strategy.launcher.launch(trainer_fn, *args, trainer=self, **kwargs)
    720             else:
--> 721                 return trainer_fn(*args, **kwargs)
    722         # TODO: treat KeyboardInterrupt as BaseException (delete the code below) in v1.7
    723         except KeyboardInterrupt as exception:

/opt/conda/lib/python3.7/site-packages/pytorch_lightning/trainer/trainer.py in _test_impl(self, model, dataloaders, ckpt_path, verbose, datamodule)
    981 
    982         # run test
--> 983         results = self._run(model, ckpt_path=self.ckpt_path)
    984 
    985         assert self.state.stopped

/opt/conda/lib/python3.7/site-packages/pytorch_lightning/trainer/trainer.py in _run(self, model, ckpt_path)
   1170         self.__setup_profiler()
   1171 
-> 1172         self._call_setup_hook()  # allow user to setup lightning_module in accelerator environment
   1173 
   1174         # check if we should delay restoring checkpoint till later

/opt/conda/lib/python3.7/site-packages/pytorch_lightning/trainer/trainer.py in _call_setup_hook(self)
   1488 
   1489         if self.datamodule is not None:
-> 1490             self.datamodule.setup(stage=fn)
   1491         self._call_callback_hooks("setup", stage=fn)
   1492         self._call_lightning_module_hook("setup", stage=fn)

/kaggle/input/matchformer/MatchFormer/model/data.py in setup(self, stage)
    114 
    115         try:
--> 116             self.world_size = dist.get_world_size()
    117             self.rank = dist.get_rank()
    118             logger.info(f"[rank:{self.rank}] world_size: {self.world_size}")

/opt/conda/lib/python3.7/site-packages/torch/distributed/distributed_c10d.py in get_world_size(group)
    865         return -1
    866 
--> 867     return _get_group_size(group)
    868 
    869 

/opt/conda/lib/python3.7/site-packages/torch/distributed/distributed_c10d.py in _get_group_size(group)
    323     """
    324     if group is GroupMember.WORLD or group is None:
--> 325         default_pg = _get_default_group()
    326         return default_pg.size()
    327     return group.size()

/opt/conda/lib/python3.7/site-packages/torch/distributed/distributed_c10d.py in _get_default_group()
    428     if not is_initialized():
    429         raise RuntimeError(
--> 430             "Default process group has not been initialized, "
    431             "please make sure to call init_process_group."
    432         )

RuntimeError: Default process group has not been initialized, please make sure to call init_process_group.

It seems to an error related to DistributedDataParallel but I am running on only 1 gpu.
Any help will be appreciated, thank you!

Bug

Hello, where did you get the data['spv_b_ids'] in coarse matching from? I've looked at the dataset file and found that there is no such key value