Thanks to visit codestin.com
Credit goes to github.com

Skip to content

Training Problem with SWIN configuration file  #225

@ronit450

Description

@ronit450

I am using mask_rcnn_swin_small_patch4_window7_mstrain_480-800_adamw_3x_coco.py, with my own training script. Its my custom dataset which I am using. I have not modified anything in the config file but getting error

import json
import mmcv
import os.path as osp
from mmdet.apis import train_detector, set_random_seed
from mmdet.datasets import build_dataset
from mmdet.models import build_detector
import mmdet
import torch
from mmdet.datasets.builder import DATASETS

class ObjectDetector:
    def __init__(self, config_path):
        torch.cuda.empty_cache()
        with open(config_path, 'r') as file:
            self.config = json.load(file)
        
        # Load the default config file
        self.cfg =  mmcv.Config.fromfile('/home/biosense/Documents/Rana/Ronit/SWIN/Swin-Transformer-Object-Detection/configs/_base_/models/cascade_mask_rcnn_r50_fpn.py')
        # self.cfg = ''
        self._update_config()
        self._save_config()

    def _update_config(self):
        # Dataset configuration
        self.cfg.dataset_type = 'CocoDataset'
        self.cfg.data_root = self.config['train_folder']

        self.cfg.data.train.type = self.cfg.dataset_type
        self.cfg.data.train.data_root = self.config['train_folder']
        self.cfg.data.train.ann_file = self.config['train_json']
        self.cfg.data.train.img_prefix = ''

        self.cfg.data.val.type = self.cfg.dataset_type
        self.cfg.data.val.data_root = self.config['val_folder']
        self.cfg.data.val.ann_file = self.config['val_json']
        self.cfg.data.val.img_prefix = ''

        self.cfg.data.test.type = self.cfg.dataset_type
        self.cfg.data.test.data_root = self.config['test_folder']
        self.cfg.data.test.ann_file = self.config['test_json']
        self.cfg.data.test.img_prefix = ''

        # Model settings
        self.cfg.model.roi_head.bbox_head.num_classes = self.config['num_classes']

        # Training settings
        self.cfg.runner.max_epochs = self.config['epochs']
        self.cfg.work_dir = './results'
        
        # Optimizer settings
        self.cfg.optimizer.lr = self.config['learning_rate']
        self.cfg.optimizer.weight_decay = self.config['weight_decay']
        
        # Learning rate scheduler settings
        self.cfg.lr_config.policy = 'step'
        self.cfg.lr_config.warmup = self.config['warmup']
        self.cfg.lr_config.warmup_iters = self.config['warmup_iters']
        self.cfg.lr_config.warmup_ratio = self.config['warmup_ratio']
        self.cfg.lr_config.step = self.config['step_lr_policy']

        # Logging settings
        self.cfg.log_config.interval = self.config['log_interval']

        # Evaluation settings
        self.cfg.evaluation.metric = 'bbox'
        self.cfg.evaluation.interval = self.config['evaluation_interval']

        # Checkpoint settings
        self.cfg.checkpoint_config.interval = self.config['checkpoint_interval']

        # Seed for reproducibility
        self.cfg.seed = 0
        set_random_seed(0, deterministic=False)
        self.cfg.gpu_ids = range(1)

        print(f'Config:\n{self.cfg.pretty_text}')

    def _save_config(self):
        # Save the updated config to a file
        self.cfg.dump('./updated_config.py')

    def train(self):
        # Build dataset and model
        mmdet.datasets.coco.CocoDataset.CLASSES = ('pt', 'gp')
        datasets = [build_dataset(self.cfg.data.train)]
        model = build_detector(self.cfg.model, train_cfg=self.cfg.get('train_cfg'), test_cfg=self.cfg.get('test_cfg'))
        model.CLASSES = datasets[0].CLASSES

        # Create directory for work
        mmcv.mkdir_or_exist(osp.abspath(self.cfg.work_dir))
        train_detector(model, datasets, self.cfg, distributed=False, validate=True)

if __name__ == '__main__':
    config_path = 'config.json'
    detector = ObjectDetector(config_path)
    detector.train()

File "/home/biosense/Documents/Rana/Ronit/SWIN/Swin-Transformer-Object-Detection/mmdet/datasets/utils.py", line 136, in _check_head
assert module.num_classes == len(dataset.CLASSES),
AssertionError: The num_classes (80) in FCNMaskHead of MMDataParallel does not matches the length of CLASSES 2) in CocoDataset

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions