Running ForAINet on basic settings

[Proeliorr]

Hello,

I am trying to run ForAInet with the default settings with as little effort as possible. I would like to use the existing pre-trained model on my datasets and have a couple of questions

1. I installed the ForAInet from the docker container using Docker file

2. I do not want to train my data, just use the pre-trained model on my dataset, so I put my dataset into the folder and adjusted the eval.yaml "data: fold..."

3. I downloaded the pre-trained model and set the checkpoint_dir:path

Here is a part of the directories where my data 

ForAinet

├──	tree_metrics
├──	superpoint_graph
├──	PointCloudSegmentation
	├──	torch_points3d
	├──	scripts
	├──	forward_scripts
	├── conf
	├── data
		├── checkpoints # I created this folder and changed it in eval.yaml file
		│	├── PointGroup-PAPER.pt #zipped model file
		│	└── archive #unzipped model file
  		│		├── data
		│		├── data.pkl
		│		└── version
		│
		├── alt12_9_eval.ply	#data for using the ForAInet - 4 .ply datasets
		├── pol01_lwpl_eval.ply
		├── Single_tree_eval.ply
		└── Wl28_FraExc_normalized_eval.ply
	├──
	... ### other files
	...

Here is the eval.yaml file:

eval.yaml

defaults: 
  - visualization: eval

num_workers: 0
batch_size: 1
cuda: 0
weight_name: "latest" # Used during resume, select with model to load from [miou, macc, acc..., latest]
enable_cudnn: True
checkpoint_dir: "/home/jakub/SmaSiKaFe/Methodologies/ForAINet/ForAINet/PointCloudSegmentation/data/checkpoints/archive"
model_name: PointGroup-PAPER
precompute_multi_scale: True # Compute multiscate features on cpu for faster training / inference
enable_dropout: False
voting_runs: 1
data: 
  # number, e.g. 3 OR ply path, e.g. "/cluster/work/igp_psr/yuayue/RA/WP1/check_jan12/TP3D_PanopticSeg/data/npm3dfused/raw/Paris.ply"
  # test files for basic settings (remember change to your data path)
  fold: ['/home/jakub/SmaSiKaFe/Methodologies/ForAINet/ForAINet/PointCloudSegmentation/data/alt12_9_eval.ply',
          '/home/jakub/SmaSiKaFe/Methodologies/ForAINet/ForAINet/PointCloudSegmentation/data/Wl28_FraExc_normalized_eval.ply',
          '/home/jakub/SmaSiKaFe/Methodologies/ForAINet/ForAINet/PointCloudSegmentation/data/pol01_lwpl_eval.ply',
          '/home/jakub/SmaSiKaFe/Methodologies/ForAINet/ForAINet/PointCloudSegmentation/data/Single_tree_eval.ply']

  # test files for understory trees experiment
  #fold: ['/cluster/work/igp_psr/binbin/OutdoorPanopticSeg_V2/data_set1_5classes/treeinsfused/raw/NIBIO2/NIBIO2_plot1_annotated_test.ply', '/cluster/work/igp_psr/binbin/OutdoorPanopticSeg_V2/data_set1_5classes/treeinsfused/raw/NIBIO2/NIBIO2_plot10_annotated_test.ply', '/cluster/work/igp_psr/binbin/OutdoorPanopticSeg_V2/data_set1_5classes/treeinsfused/raw/NIBIO2/NIBIO2_plot15_annotated_test.ply', '/cluster/work/igp_psr/binbin/OutdoorPanopticSeg_V2/data_set1_5classes/treeinsfused/raw/NIBIO2/NIBIO2_plot27_annotated_test.ply', '/cluster/work/igp_psr/binbin/OutdoorPanopticSeg_V2/data_set1_5classes/treeinsfused/raw/NIBIO2/NIBIO2_plot3_annotated_test.ply', '/cluster/work/igp_psr/binbin/OutdoorPanopticSeg_V2/data_set1_5classes/treeinsfused/raw/NIBIO2/NIBIO2_plot32_annotated_test.ply', '/cluster/work/igp_psr/binbin/OutdoorPanopticSeg_V2/data_set1_5classes/treeinsfused/raw/NIBIO2/NIBIO2_plot34_annotated_test.ply', '/cluster/work/igp_psr/binbin/OutdoorPanopticSeg_V2/data_set1_5classes/treeinsfused/raw/NIBIO2/NIBIO2_plot35_annotated_test.ply', '/cluster/work/igp_psr/binbin/OutdoorPanopticSeg_V2/data_set1_5classes/treeinsfused/raw/NIBIO2/NIBIO2_plot48_annotated_test.ply', '/cluster/work/igp_psr/binbin/OutdoorPanopticSeg_V2/data_set1_5classes/treeinsfused/raw/NIBIO2/NIBIO2_plot49_annotated_test.ply', '/cluster/work/igp_psr/binbin/OutdoorPanopticSeg_V2/data_set1_5classes/treeinsfused/raw/NIBIO2/NIBIO2_plot52_annotated_test.ply', '/cluster/work/igp_psr/binbin/OutdoorPanopticSeg_V2/data_set1_5classes/treeinsfused/raw/NIBIO2/NIBIO2_plot53_annotated_test.ply', '/cluster/work/igp_psr/binbin/OutdoorPanopticSeg_V2/data_set1_5classes/treeinsfused/raw/NIBIO2/NIBIO2_plot58_annotated_test.ply', '/cluster/work/igp_psr/binbin/OutdoorPanopticSeg_V2/data_set1_5classes/treeinsfused/raw/NIBIO2/NIBIO2_plot6_annotated_test.ply', '/cluster/work/igp_psr/binbin/OutdoorPanopticSeg_V2/data_set1_5classes/treeinsfused/raw/NIBIO2/NIBIO2_plot60_annotated_test.ply']

tracker_options: # Extra options for the tracker
  full_res: True
  make_submission: True
  ply_output: "vote1regular.ply"

hydra:
  run:
    dir: ${checkpoint_dir}/eval/${now:%Y-%m-%d_%H-%M-%S}

However, after running:

cd /forainet/PointCloudSegmentation
python3 eval.py

the Value error occurs.

Traceback (most recent call last):
  File "eval.py", line 22, in <module>
    main()
  File "/usr/local/lib/python3.8/dist-packages/hydra/main.py", line 32, in decorated_main
    _run_hydra(
  File "/usr/local/lib/python3.8/dist-packages/hydra/_internal/utils.py", line 346, in _run_hydra
    run_and_report(
  File "/usr/local/lib/python3.8/dist-packages/hydra/_internal/utils.py", line 201, in run_and_report
    raise ex
  File "/usr/local/lib/python3.8/dist-packages/hydra/_internal/utils.py", line 198, in run_and_report
    return func()
  File "/usr/local/lib/python3.8/dist-packages/hydra/_internal/utils.py", line 347, in <lambda>
    lambda: hydra.run(
  File "/usr/local/lib/python3.8/dist-packages/hydra/_internal/hydra.py", line 107, in run
    return run_job(
  File "/usr/local/lib/python3.8/dist-packages/hydra/core/utils.py", line 129, in run_job
    ret.return_value = task_function(task_cfg)
  File "eval.py", line 13, in main
    trainer = Trainer(cfg)
  File "/app/forainet/PointCloudSegmentation/torch_points3d/trainer.py", line 48, in __init__
    self._initialize_trainer()
  File "/app/forainet/PointCloudSegmentation/torch_points3d/trainer.py", line 80, in _initialize_trainer
    self._checkpoint: ModelCheckpoint = ModelCheckpoint(
  File "/app/forainet/PointCloudSegmentation/torch_points3d/metrics/model_checkpoint.py", line 173, in __init__
    self._checkpoint = Checkpoint.load(load_dir, check_name, run_config=rc, strict=strict, resume=resume)
  File "/app/forainet/PointCloudSegmentation/torch_points3d/metrics/model_checkpoint.py", line 73, in load
    raise ValueError(message)

ValueError: The provided path /home/jakub/SmaSiKaFe/Methodologies/ForAINet/ForAINet/PointCloudSegmentation/data/checkpoints/PointGroup-PAPER.pt didn't contain the checkpoint_file PointGroup-PAPER.pt

Are there any other .yaml files that I must change or correct to make it work?

[CiSong10]

Hope you’re able to figure it out! I’m also planning to try deploying ForAINet directly first, then experiment with fine-tuning. The tutorial in the README.md is a bit brief at the moment, and as someone who's not very experienced with deep learning, I am struggling to follow it. I hope a more detailed tutorial will be available in the future.

[Proeliorr]

@CiSong10 I agree.
Maybe that will be a hint for you: If you want run the ForAInet to to get familiar with results on your datasets, the ForestSens platform has it embedded as on of its algorithms. That could be your first approach, but at some point you will need to have it done in a "proper" way.

[CiSong10]

@Proeliorr Thanks for your tips! I did manage to run ForAINet within ForestSens, but I’m now trying to deploy it directly and running into some new issues. That said, I think I can help with yours:

The ValueError you’re seeing says: The provided path /home/jakub/SmaSiKaFe/Methodologies/ForAINet/ForAINet/PointCloudSegmentation/data/checkpoints/PointGroup-PAPER.pt didn't contain the checkpoint_file PointGroup-PAPER.pt

It looks like the script is interpreting .../data/checkpoints/PointGroup-PAPER.pt as a directory, but it's not. So it fails when trying to find the checkpoint_file inside it.

Specifically, check line 7 of your eval.py , where you likely have:

@hydra.main(config_path="conf", config_name="eval")

This tells Hydra to load the config file from conf/eval.yaml. Can you confirm that line is present? And in your YAML config, make sure the checkpoint_dir parameter points to the directory containing PointGroup-PAPER.

Let me know if that helps!

[Proeliorr]

@CiSong10. Thanks for your answer.

1. The eval.py looks like this:

import hydra
from hydra.core.global_hydra import GlobalHydra
from omegaconf import OmegaConf
from torch_points3d.trainer import Trainer


@hydra.main(config_path="conf", config_name="eval")
def main(cfg):
    OmegaConf.set_struct(cfg, False)  # This allows getattr and hasattr methods to function correctly
    if cfg.pretty_print:
        print(OmegaConf.to_yaml(cfg))
    
    trainer = Trainer(cfg)
    trainer.eval(stage_name = "test")
    #
    # # https://github.com/facebookresearch/hydra/issues/440
    GlobalHydra.get_state().clear()
    return 0


if __name__ == "__main__":
    main()

So, as you can see, the line mentioned by you is present.

2. The eval.yaml looks like this:

defaults: 
  - visualization: eval

num_workers: 0
batch_size: 1
cuda: 0
weight_name: 'latest' # Used during resume, select with model to load from [miou, macc, acc..., latest]
enable_cudnn: True
#checkpoint_dir: # '{your_output_log_path}/outputs/2020-01-28/11-04-13'
# checkpoint_dir: '/home/jakub/SmaSiKaFe/Methodologies/ForAINet/ForAINet/PointCloudSegmentation/data/checkpoints/PointGroup-PAPER.pt'
checkpoint_dir: '/home/jakub/SmaSiKaFe/Methodologies/ForAINet/ForAINet/PointCloudSegmentation/data/checkpoints/'
model_name: PointGroup-PAPER
precompute_multi_scale: True # Compute multiscale features on cpu for faster training / inference
enable_dropout: False
voting_runs: 1
data: 
  # number, e.g. 3 OR ply path, e.g. '/cluster/work/igp_psr/yuayue/RA/WP1/check_jan12/TP3D_PanopticSeg/data/npm3dfused/raw/Paris.ply'
  # test files for basic settings (remember to change to your data path)
  fold: ['/home/jakub/SmaSiKaFe/Methodologies/ForAINet/ForAINet/PointCloudSegmentation/data/alt12_9_eval.ply',
          '/home/jakub/SmaSiKaFe/Methodologies/ForAINet/ForAINet/PointCloudSegmentation/data/Wl28_FraExc_normalized_eval.ply',
          '/home/jakub/SmaSiKaFe/Methodologies/ForAINet/ForAINet/PointCloudSegmentation/data/pol01_lwpl_eval.ply',
          '/home/jakub/SmaSiKaFe/Methodologies/ForAINet/ForAINet/PointCloudSegmentation/data/Single_tree_eval.ply']
  # test files for understory trees experiment
tracker_options: # Extra options for the tracker
  full_res: True
  make_submission: True
  ply_output: 'vote1regular.ply'

hydra:
  run:
    dir: ${checkpoint_dir}/eval/${now:%Y-%m-%d_%H-%M-%S}

and the checkpoint_dir is set correctly, too.

The given checkpoint folder contains:

where archive is the unzipped PointGroup-PAPER.pt file (I was not sure if I should unzip it or not).

So, I believe all necessary adjustments have already been made, and I cannot really find where a possible error (errors) could be hidden.

[CiSong10]

@Proeliorr yeah, this looks good. Can you try run it again? Do you still have the same error message?

[Proeliorr]

Yes, I run the eval.py once again and the same error occurred.
The whole error message:

root@02c5cea00c20:/app/forainet/PointCloudSegmentation# python3 eval.py
/usr/local/lib/python3.8/dist-packages/requests/__init__.py:102: RequestsDependencyWarning: urllib3 (1.26.7) or chardet (5.2.0)/charset_normalizer (2.0.6) doesn't match a supported version!
  warnings.warn("urllib3 ({}) or chardet ({})/charset_normalizer ({}) doesn't match a supported "
[2025-06-12 13:37:56,926][torch_points3d.trainer][INFO] - DEVICE : cuda
Traceback (most recent call last):
  File "eval.py", line 13, in main
    trainer = Trainer(cfg)
  File "/app/forainet/PointCloudSegmentation/torch_points3d/trainer.py", line 48, in __init__
    self._initialize_trainer()
  File "/app/forainet/PointCloudSegmentation/torch_points3d/trainer.py", line 80, in _initialize_trainer
    self._checkpoint: ModelCheckpoint = ModelCheckpoint(
  File "/app/forainet/PointCloudSegmentation/torch_points3d/metrics/model_checkpoint.py", line 173, in __init__
    self._checkpoint = Checkpoint.load(load_dir, check_name, run_config=rc, strict=strict, resume=resume)
  File "/app/forainet/PointCloudSegmentation/torch_points3d/metrics/model_checkpoint.py", line 79, in load
    shutil.copyfile(
  File "/usr/lib/python3.8/shutil.py", line 264, in copyfile
    with open(src, 'rb') as fsrc, open(dst, 'wb') as fdst:
IsADirectoryError: [Errno 21] Is a directory: '/home/jakub/SmaSiKaFe/Methodologies/ForAINet/ForAINet/PointCloudSegmentation/data/checkpoints/PointGroup-PAPER.pt'

Set the environment variable HYDRA_FULL_ERROR=1 for a complete stack trace.

and after setting HYDRA_FULL_ERROR=1

/usr/local/lib/python3.8/dist-packages/requests/__init__.py:102: RequestsDependencyWarning: urllib3 (1.26.7) or chardet (5.2.0)/charset_normalizer (2.0.6) doesn't match a supported version!
  warnings.warn("urllib3 ({}) or chardet ({})/charset_normalizer ({}) doesn't match a supported "
[2025-06-12 13:39:56,276][torch_points3d.trainer][INFO] - DEVICE : cuda
Traceback (most recent call last):
  File "eval.py", line 22, in <module>
    main()
  File "/usr/local/lib/python3.8/dist-packages/hydra/main.py", line 32, in decorated_main
    _run_hydra(
  File "/usr/local/lib/python3.8/dist-packages/hydra/_internal/utils.py", line 346, in _run_hydra
    run_and_report(
  File "/usr/local/lib/python3.8/dist-packages/hydra/_internal/utils.py", line 201, in run_and_report
    raise ex
  File "/usr/local/lib/python3.8/dist-packages/hydra/_internal/utils.py", line 198, in run_and_report
    return func()
  File "/usr/local/lib/python3.8/dist-packages/hydra/_internal/utils.py", line 347, in <lambda>
    lambda: hydra.run(
  File "/usr/local/lib/python3.8/dist-packages/hydra/_internal/hydra.py", line 107, in run
    return run_job(
  File "/usr/local/lib/python3.8/dist-packages/hydra/core/utils.py", line 129, in run_job
    ret.return_value = task_function(task_cfg)
  File "eval.py", line 13, in main
    trainer = Trainer(cfg)
  File "/app/forainet/PointCloudSegmentation/torch_points3d/trainer.py", line 48, in __init__
    self._initialize_trainer()
  File "/app/forainet/PointCloudSegmentation/torch_points3d/trainer.py", line 80, in _initialize_trainer
    self._checkpoint: ModelCheckpoint = ModelCheckpoint(
  File "/app/forainet/PointCloudSegmentation/torch_points3d/metrics/model_checkpoint.py", line 173, in __init__
    self._checkpoint = Checkpoint.load(load_dir, check_name, run_config=rc, strict=strict, resume=resume)
  File "/app/forainet/PointCloudSegmentation/torch_points3d/metrics/model_checkpoint.py", line 79, in load
    shutil.copyfile(
  File "/usr/lib/python3.8/shutil.py", line 264, in copyfile
    with open(src, 'rb') as fsrc, open(dst, 'wb') as fdst:
IsADirectoryError: [Errno 21] Is a directory: '/home/jakub/SmaSiKaFe/Methodologies/ForAINet/ForAINet/PointCloudSegmentation/data/checkpoints/PointGroup-PAPER.pt'

I assume this can be an issue here: I mean this incompatibility

RequestsDependencyWarning: urllib3 (1.26.7) or chardet (5.2.0)/charset_normalizer (2.0.6) doesn't match a supported version! warnings.warn("urllib3 ({}) or chardet ({})/charset_normalizer ({}) doesn't match a supported " [2025-06-12 13:39:56,276][torch_points3d.trainer][INFO] - DEVICE : cuda

I think that updating the Dockerfile itself could be a potential solution.

[CiSong10]

Hi @Proeliorr , actually — this is a different issue. Your original ValueError is solved — yay!

This IsADirectoryError suggests that your checkpoints/PointGroup-PAPER.pt is a directory, not a file. This is weird. Can you check if your checkpoints/PointGroup-PAPER.pt is actually a folder? Make sure you’re pointing to the unzipped model file you downloaded.

As for the compatibility warning — it’s unrelated to this error. I get the same one. It happens because chardet==5.2.0 was pulled in during the Docker build, but it's incompatible with requests==2.26.0. You can fix it by uninstalling chardet inside the container. I’m also updating the Dockerfile to fix this automatically — I’ll upload it once I’ve confirmed it works.

[Proeliorr]

1. The screenshot from my previous comment showed that the PointGroup-PAPER.pt is a zip archive.

However, the PT extension of this file means that it is a default extension for saving the models by PyTorch. The given links also provide instructions for loading Torch models (also with .pt extension), and the ForAINet instruction does not mention anything that the model should be unzipped (which could be obvious, too). Anyway, it would be pretty inconsistent to export sth as a model file and load it in the unzipped format.

2. I run the command after unzipping the .PT file and specifying the "archive" folder in the eval.yml file.

root@02c5cea00c20:/app/forainet/PointCloudSegmentation# python3 eval.py
/usr/local/lib/python3.8/dist-packages/requests/__init__.py:102: RequestsDependencyWarning: urllib3 (1.26.7) or chardet (5.2.0)/charset_normalizer (2.0.6) doesn't match a supported version!
  warnings.warn("urllib3 ({}) or chardet ({})/charset_normalizer ({}) doesn't match a supported "
[2025-06-12 14:46:59,271][torch_points3d.trainer][INFO] - DEVICE : cuda
Traceback (most recent call last):
  File "eval.py", line 22, in <module>
    main()
  File "/usr/local/lib/python3.8/dist-packages/hydra/main.py", line 32, in decorated_main
    _run_hydra(
  File "/usr/local/lib/python3.8/dist-packages/hydra/_internal/utils.py", line 346, in _run_hydra
    run_and_report(
  File "/usr/local/lib/python3.8/dist-packages/hydra/_internal/utils.py", line 201, in run_and_report
    raise ex
  File "/usr/local/lib/python3.8/dist-packages/hydra/_internal/utils.py", line 198, in run_and_report
    return func()
  File "/usr/local/lib/python3.8/dist-packages/hydra/_internal/utils.py", line 347, in <lambda>
    lambda: hydra.run(
  File "/usr/local/lib/python3.8/dist-packages/hydra/_internal/hydra.py", line 107, in run
    return run_job(
  File "/usr/local/lib/python3.8/dist-packages/hydra/core/utils.py", line 129, in run_job
    ret.return_value = task_function(task_cfg)
  File "eval.py", line 13, in main
    trainer = Trainer(cfg)
  File "/app/forainet/PointCloudSegmentation/torch_points3d/trainer.py", line 48, in __init__
    self._initialize_trainer()
  File "/app/forainet/PointCloudSegmentation/torch_points3d/trainer.py", line 80, in _initialize_trainer
    self._checkpoint: ModelCheckpoint = ModelCheckpoint(
  File "/app/forainet/PointCloudSegmentation/torch_points3d/metrics/model_checkpoint.py", line 173, in __init__
    self._checkpoint = Checkpoint.load(load_dir, check_name, run_config=rc, strict=strict, resume=resume)
  File "/app/forainet/PointCloudSegmentation/torch_points3d/metrics/model_checkpoint.py", line 79, in load
    shutil.copyfile(
  File "/usr/lib/python3.8/shutil.py", line 264, in copyfile
    with open(src, 'rb') as fsrc, open(dst, 'wb') as fdst:
IsADirectoryError: [Errno 21] Is a directory: '/home/jakub/SmaSiKaFe/Methodologies/ForAINet/ForAINet/PointCloudSegmentation/data/checkpoints/archive/PointGroup-PAPER.pt'
root@02c5cea00c20:/app/forainet/PointCloudSegmentation# root@02c5cea00c20:/app/forainet/PointCloudSegmentation# python3 eval.py
/usr/local/lib/python3.8/dist-packages/requests/__init__.py:102: RequestsDependencyWarning: urllib3 (1.26.7) or chardet (5.2.0)/charset_normalizer (2.0.6) doesn't match a supported version!
  warnings.warn("urllib3 ({}) or chardet ({})/charset_normalizer ({}) doesn't match a supported "
[2025-06-12 14:46:59,271][torch_points3d.trainer][INFO] - DEVICE : cuda
Traceback (most recent call last):
  File "eval.py", line 22, in <module>
    main()
  File "/usr/local/lib/python3.8/dist-packages/hydra/main.py", line 32, in decorated_main
    _run_hydra(
  File "/usr/local/lib/python3.8/dist-packages/hydra/_internal/utils.py", line 346, in _run_hydra
    run_and_report(
  File "/usr/local/lib/python3.8/dist-packages/hydra/_internal/utils.py", line 201, in run_and_report
    raise ex
  File "/usr/local/lib/python3.8/dist-packages/hydra/_internal/utils.py", line 198, in run_and_report
    return func()
  File "/usr/local/lib/python3.8/dist-packages/hydra/_internal/utils.py", line 347, in <lambda>
    lambda: hydra.run(
  File "/usr/local/lib/python3.8/dist-packages/hydra/_internal/hydra.py", line 107, in run
    return run_job(
  File "/usr/local/lib/python3.8/dist-packages/hydra/core/utils.py", line 129, in run_job
    ret.return_value = task_function(task_cfg)
  File "eval.py", line 13, in main
    trainer = Trainer(cfg)
  File "/app/forainet/PointCloudSegmentation/torch_points3d/trainer.py", line 48, in __init__
    self._initialize_trainer()
  File "/app/forainet/PointCloudSegmentation/torch_points3d/trainer.py", line 80, in _initialize_trainer
    self._checkpoint: ModelCheckpoint = ModelCheckpoint(
  File "/app/forainet/PointCloudSegmentation/torch_points3d/metrics/model_checkpoint.py", line 173, in __init__
    self._checkpoint = Checkpoint.load(load_dir, check_name, run_config=rc, strict=strict, resume=resume)
  File "/app/forainet/PointCloudSegmentation/torch_points3d/metrics/model_checkpoint.py", line 79, in load
    shutil.copyfile(
  File "/usr/lib/python3.8/shutil.py", line 264, in copyfile
    with open(src, 'rb') as fsrc, open(dst, 'wb') as fdst:
IsADirectoryError: [Errno 21] Is a directory: '/home/jakub/SmaSiKaFe/Methodologies/ForAINet/ForAINet/PointCloudSegmentation/data/checkpoints/archive/PointGroup-PAPER.pt'

It appears that the eval.py or another file looks for the PointGroup-PAPER.pt name, specifically.
IsADirectoryError: [Errno 21] Is a directory: '/home/jakub/SmaSiKaFe/Methodologies/ForAINet/ForAINet/PointCloudSegmentation/data/checkpoints/archive/PointGroup-PAPER.pt'

PS: Your meme is even more actual than it has been before :)

[CiSong10]

I re-downloaded PointGroup-PAPER.pt , and it’s not an archive on my end—just a regular file. Maybe try re-downloading it, or try unzipping it?

Also, I updated the Dockerfile to fix the compatibility warning. Hope that helps! I'm also working on a note to document how to run the model, but it’s still pretty empty since I haven’t fully figured things out yet 😅.

Currently I have this issue (only 10 lines ahead of you in trainer.py 😅) :

root@0e134f6cfc8f:/home/cisong/ForAINet# python3 PointCloudSegmentation/predict.py
[2025-06-12 10:29:17,803][torch_points3d.trainer][INFO] - DEVICE : cuda
[2025-06-12 10:29:19,006][torch_points3d.metrics.model_checkpoint][INFO] - Loading checkpoint from /home/cisong/ForAINet/outputs/pretrained/PointGroup-PAPER.pt
Traceback (most recent call last):
  File "PointCloudSegmentation/predict.py", line 13, in main
    trainer = Trainer(cfg)
  File "/home/cisong/ForAINet/PointCloudSegmentation/torch_points3d/trainer.py", line 48, in __init__
    self._initialize_trainer()
  File "/home/cisong/ForAINet/PointCloudSegmentation/torch_points3d/trainer.py", line 90, in _initialize_trainer
    self._dataset: BaseDataset = instantiate_dataset(self._checkpoint.data_config)
  File "/home/cisong/ForAINet/PointCloudSegmentation/torch_points3d/datasets/dataset_factory.py", line 46, in instantiate_dataset
    dataset = dataset_cls(dataset_config)
  File "/home/cisong/ForAINet/PointCloudSegmentation/torch_points3d/datasets/panoptic/treeins_set1.py", line 633, in __init__
    self.test_dataset = dataset_cls(
  File "/home/cisong/ForAINet/PointCloudSegmentation/torch_points3d/datasets/segmentation/treeins_set1.py", line 599, in __init__
    super().__init__(root, grid_size, *args, **kwargs)
  File "/home/cisong/ForAINet/PointCloudSegmentation/torch_points3d/datasets/segmentation/treeins_set1.py", line 239, in __init__
    super(TreeinsOriginalFused, self).__init__(root, transform, pre_transform, pre_filter)
  File "/usr/local/lib/python3.8/dist-packages/torch_geometric/data/in_memory_dataset.py", line 60, in __init__
    super().__init__(root, transform, pre_transform, pre_filter)
  File "/usr/local/lib/python3.8/dist-packages/torch_geometric/data/dataset.py", line 83, in __init__
    self._download()
  File "/usr/local/lib/python3.8/dist-packages/torch_geometric/data/dataset.py", line 140, in _download
    self.download()
  File "/home/cisong/ForAINet/PointCloudSegmentation/torch_points3d/datasets/panoptic/treeins_set1.py", line 560, in download
    super().download()
  File "/home/cisong/ForAINet/PointCloudSegmentation/torch_points3d/datasets/segmentation/treeins_set1.py", line 625, in download
    super().download()
  File "/usr/local/lib/python3.8/dist-packages/torch_geometric/data/in_memory_dataset.py", line 50, in download
    raise NotImplementedError
NotImplementedError

My gut feeling is that these scripts were primarily written for datasets with ground truth labels, so a lot of the code assumes the presence of test/eval data. I'm planning to go through the scripts and try modifying them to work with raw point cloud data instead.

@bxiang233, I'd really appreciate it if you could confirm whether that's the case and share any guidance on how to deploy the model on raw data. I noticed you included a workflow for tiling, predicting, and merging in the README, and—if I understand correctly—that process still relies on eval.py.

I also wish I knew how the ForestSens team handled this.

[Proeliorr]

1. In my case, running your Docker file fixed the Warnings. However, it has not repaired the main issue:

oot@1163dffa2cf8:/app/forainet_song/PointCloudSegmentation# python3 eval.py
[2025-06-13 08:20:41,933][torch_points3d.trainer][INFO] - DEVICE : cuda
Traceback (most recent call last):
  File "eval.py", line 22, in <module>
    main()
  File "/usr/local/lib/python3.8/dist-packages/hydra/main.py", line 32, in decorated_main
    _run_hydra(
  File "/usr/local/lib/python3.8/dist-packages/hydra/_internal/utils.py", line 346, in _run_hydra
    run_and_report(
  File "/usr/local/lib/python3.8/dist-packages/hydra/_internal/utils.py", line 201, in run_and_report
    raise ex
  File "/usr/local/lib/python3.8/dist-packages/hydra/_internal/utils.py", line 198, in run_and_report
    return func()
  File "/usr/local/lib/python3.8/dist-packages/hydra/_internal/utils.py", line 347, in <lambda>
    lambda: hydra.run(
  File "/usr/local/lib/python3.8/dist-packages/hydra/_internal/hydra.py", line 107, in run
    return run_job(
  File "/usr/local/lib/python3.8/dist-packages/hydra/core/utils.py", line 129, in run_job
    ret.return_value = task_function(task_cfg)
  File "eval.py", line 13, in main
    trainer = Trainer(cfg)
  File "/app/forainet_song/PointCloudSegmentation/torch_points3d/trainer.py", line 48, in __init__
    self._initialize_trainer()
  File "/app/forainet_song/PointCloudSegmentation/torch_points3d/trainer.py", line 80, in _initialize_trainer
    self._checkpoint: ModelCheckpoint = ModelCheckpoint(
  File "/app/forainet_song/PointCloudSegmentation/torch_points3d/metrics/model_checkpoint.py", line 173, in __init__
    self._checkpoint = Checkpoint.load(load_dir, check_name, run_config=rc, strict=strict, resume=resume)
  File "/app/forainet_song/PointCloudSegmentation/torch_points3d/metrics/model_checkpoint.py", line 73, in load
    raise ValueError(message)
ValueError: The provided path /home/jakub/SmaSiKaFe/Methodologies/ForAINet/ForAINet/PointCloudSegmentation/data/checkpoints/ didn't contain the checkpoint_file PointGroup-PAPER.pt

ValueError: The provided path /home/jakub/SmaSiKaFe/Methodologies/ForAINet/ForAINet/PointCloudSegmentation/data/checkpoints/ didn't contain the checkpoint_file PointGroup-PAPER.pt

I checked the correctness of the data paths more than once, but it may be related to the Docker pathing itself. I mounted the ForAINet data to the /app folder inside the container, which may not be interpreted by the container properly. I changed the parts of filepaths from /home/jakub/SmaSiKaFe/Methodologies/ForAINet/ForAINet/* to /app/forainet_song/PointCloudSegmentation/* and received this error, so perhaps the previous error has been solved:

[2025-06-13 08:41:29,160][torch_points3d.trainer][INFO] - DEVICE : cuda
[2025-06-13 08:41:29,647][torch_points3d.metrics.model_checkpoint][INFO] - Loading checkpoint from /app/forainet_song/PointCloudSegmentation/data/checkpoints/PointGroup-PAPER.pt
Traceback (most recent call last):
  File "eval.py", line 22, in <module>
    main()
  File "/usr/local/lib/python3.8/dist-packages/hydra/main.py", line 32, in decorated_main
    _run_hydra(
  File "/usr/local/lib/python3.8/dist-packages/hydra/_internal/utils.py", line 346, in _run_hydra
    run_and_report(
  File "/usr/local/lib/python3.8/dist-packages/hydra/_internal/utils.py", line 201, in run_and_report
    raise ex
  File "/usr/local/lib/python3.8/dist-packages/hydra/_internal/utils.py", line 198, in run_and_report
    return func()
  File "/usr/local/lib/python3.8/dist-packages/hydra/_internal/utils.py", line 347, in <lambda>
    lambda: hydra.run(
  File "/usr/local/lib/python3.8/dist-packages/hydra/_internal/hydra.py", line 107, in run
    return run_job(
  File "/usr/local/lib/python3.8/dist-packages/hydra/core/utils.py", line 129, in run_job
    ret.return_value = task_function(task_cfg)
  File "eval.py", line 13, in main
    trainer = Trainer(cfg)
  File "/app/forainet_song/PointCloudSegmentation/torch_points3d/trainer.py", line 48, in __init__
    self._initialize_trainer()
  File "/app/forainet_song/PointCloudSegmentation/torch_points3d/trainer.py", line 90, in _initialize_trainer
    self._dataset: BaseDataset = instantiate_dataset(self._checkpoint.data_config)
  File "/app/forainet_song/PointCloudSegmentation/torch_points3d/datasets/dataset_factory.py", line 46, in instantiate_dataset
    dataset = dataset_cls(dataset_config)
  File "/app/forainet_song/PointCloudSegmentation/torch_points3d/datasets/panoptic/treeins_set1.py", line 633, in __init__
    self.test_dataset = dataset_cls(
  File "/app/forainet_song/PointCloudSegmentation/torch_points3d/datasets/segmentation/treeins_set1.py", line 599, in __init__
    super().__init__(root, grid_size, *args, **kwargs)
  File "/app/forainet_song/PointCloudSegmentation/torch_points3d/datasets/segmentation/treeins_set1.py", line 239, in __init__
    super(TreeinsOriginalFused, self).__init__(root, transform, pre_transform, pre_filter)
  File "/usr/local/lib/python3.8/dist-packages/torch_geometric/data/in_memory_dataset.py", line 60, in __init__
    super().__init__(root, transform, pre_transform, pre_filter)
  File "/usr/local/lib/python3.8/dist-packages/torch_geometric/data/dataset.py", line 83, in __init__
    self._download()
  File "/usr/local/lib/python3.8/dist-packages/torch_geometric/data/dataset.py", line 140, in _download
    self.download()
  File "/app/forainet_song/PointCloudSegmentation/torch_points3d/datasets/panoptic/treeins_set1.py", line 560, in download
    super().download()
  File "/app/forainet_song/PointCloudSegmentation/torch_points3d/datasets/segmentation/treeins_set1.py", line 625, in download
    super().download()
  File "/usr/local/lib/python3.8/dist-packages/torch_geometric/data/in_memory_dataset.py", line 50, in download
    raise NotImplementedError
NotImplementedError

which appears to be the same issue you created in #22.

2. Regarding raw data input: In one of the previous emails to @bxiang233 , I formulated a question:
   Having the pre-trained model (as mentioned above) allows me to assume that I can skip the training phase. Am I correct?

and received: Pre-trained model – yes, skip training. Could we assume from a given cited text that the raw data is the default format? Otherwise, @bxiang233 should have mentioned that in her answer....? In addition, I contacted ForestSens authors via their email on the project website a couple of weeks ago, but haven't received any answer yet. I will let you know about any response from them — hopefully it will come in some time.

[CiSong10]

1. OMG yes that was the issue all the time. When you run the code in the Docker container, you need to point to the file path of the container not the host machine. Glad that finally solved.
2. Yes this NotImplementedError is the same I mentioned in issue 22. That seems to be somehow related to where to put the data as well. I haven't fully figure it out. I will update there when I do.

[Proeliorr]

1. Yea, I was wondering why the file is not visible in certain location despite being there, so I was going through the filepaths in docker shell and noticed this uncertainity. But I am satisfied that another error has been fixed.

2. The friend of mine knows one of the authors of ForestSens and promised me to contact him. Maybe it will accelerate the response.

[bxiang233]

Hi guys, sorry for the late reply and thank you for your discussions here. Yes, the file system inside the Docker container is totally separate from your host machine. A file path that works on your host will not work inside the container unless you properly mount the directory when starting the container.

maybe you could try:
docker run -v /path/on/host:/path/in/container ...
And when writing code inside Docker, always check and use the path as seen inside the container.

[Proeliorr]

Hello @bxiang233. Of course I run given command with the -v switch. Otherwise I would not have the access to the container files from the detached command line -d switch.
I run the docker command:

docker run -itd --gpus all --name ForAINet -v /home/path/to/a/host/directory/where/a/docker/file/is/located:/app/ForAINet ForAINet

and changed the data inputs paths, too (look at the previous comment of mine). After changing the filepaths for compatible for Docker container I encountered exact same problem as in the #22. I am patiently waiting for the solution and am available to help.

[bxiang233]

Hello @bxiang233. Of course I run given command with the -v switch. Otherwise I would not have the access to the container files from the detached command line -d switch. I run the docker command:

docker run -itd --gpus all --name ForAINet -v /home/path/to/a/host/directory/where/a/docker/file/is/located:/app/ForAINet ForAINet

and changed the data inputs paths, too (look at the previous comment of mine). After changing the filepaths for compatible for Docker container I encountered exact same problem as in the #22. I am patiently waiting for the solution and am available to help.

Yes, by default it requires the ground truth values. You can assign dummy labels (e.g., 0), which won’t affect the prediction itself. The only thing is that the evaluation metrics won’t be meaningful in that case.