new picai seg #310
new picai seg #310
Conversation
| run: | ||
| running_modes : ['train', 'infer', 'eval'] | ||
| train: | ||
| target : seg3d |
There was a problem hiding this comment.
would be useful to add comments describing the meaning and possible values of "non obvious" params
There was a problem hiding this comment.
added comments in hydra on non trivial params, will also add readme once we finalize the code
| batch_size: 16 | ||
| learning_rate : 1e-3 | ||
| weight_decay : 0 | ||
| run_sample : 0 |
There was a problem hiding this comment.
it is to limit the dataset size,
I use it for the test ( instead of saving "golden" version like Moshiko used to do)
| total_mask_prob: 0 | ||
| loss_config : | ||
| segmentation_lambda : 0.2 | ||
| classification_lamba : 0.8 |
There was a problem hiding this comment.
those params will be removed, it is from old version actually
| from fuse.data.utils.split import dataset_balanced_division_to_folds | ||
| from fuse.dl.losses.loss_default import LossDefault | ||
|
|
||
| # from report_guided_annotation import extract_lesion_candidates |
There was a problem hiding this comment.
I still keep it because maybe it will turn to detection task and then I need those imports
|
|
||
| # assert ( | ||
| # "PICAI_DATA_PATH" in os.environ | ||
| # ), "Expecting environment variable CMMD_DATA_PATH to be set. Follow the instruction in example README file to download and set the path to the data" |
There was a problem hiding this comment.
you mean PICAI_DATA_PATH, no?
| if "infer" in cfg["run.running_modes"]: | ||
| run_infer(NDict(cfg["infer"]), NDict(cfg["paths"]), NDict(cfg["train"])) | ||
| # | ||
| # analyze - skipping as it crushes without metrics |
There was a problem hiding this comment.
so why leave the run_eval function?
There was a problem hiding this comment.
same as above, will try to convert to either of the 2 solutions.
There was a problem hiding this comment.
You have metric now, so you can delete this comment.
fuse_examples/tests/conf/config.yaml
Outdated
| trainer: | ||
| accelerator : gpu | ||
| devices : 1 | ||
| num_epochs : 100 |
There was a problem hiding this comment.
you do 100 epochs just for a unit test?
There was a problem hiding this comment.
I actually override it in the test,
I will try to read the same config file as in regular run instead of creating a test config
| cfg = NDict(hydra.utils.instantiate(cfg)) | ||
| cfg["paths.working_dir"] = root | ||
| cfg["train.run_sample"] = 100 | ||
| cfg["train.trainer.num_epochs"] = 1 |
| sample_dict[key_out + seq] = image_data | ||
| return sample_dict | ||
| except: | ||
| return None |
There was a problem hiding this comment.
maybe you should display a clearer warning when this happens?
There was a problem hiding this comment.
will add a print error message before return None
fuseimg/datasets/picai.py
Outdated
| super().__init__(**kwargs) | ||
| self._dir_path = dir_path | ||
| self._data_dir = data_dir | ||
| # self._sequences = seuqences |
No description provided.