objectrelator/model/mask_decoder/Mask2Former_Simplify/main.py

#!/usr/bin/env python
# -*- encoding: utf-8 -*-
'''
@File    :   main.py
@Time    :   2022/10/11 19:54:03
@Author  :   zzubqh 
@Version :   1.0
@Contact :   baiqh@microport.com
@License :   (C)Copyright 2017-2018, Liugroup-NLPR-CASIA
@Desc    :   None
'''

# here put the import lib

import argparse
import os
os.environ['CUDA_VISIBLE_DEVICES']='0,1,2,3,4,5,6,7'  # 

from fvcore.common.config import CfgNode
from configs.config import Config
import torch
from maskformer_train import MaskFormer
from dataset.dataset import ADE200kDataset, NuImagesDataset
from Segmentation import Segmentation

if torch.cuda.device_count() > 1:
    torch.distributed.init_process_group(backend='nccl')

def user_scattered_collate(batch):
    data = [item['images'] for item in batch]
    masks = [item['masks'] for item in batch]
    out = {'images': torch.cat(data, dim=0), 'masks': torch.cat(masks, dim=0)}
    return out

def get_args():
    parser = argparse.ArgumentParser()
    parser.add_argument('--config', type=str, default='configs/maskformer_nuimages.yaml')
    parser.add_argument('--local_rank', type=int, default=0)
    parser.add_argument("--ngpus", default=1, type=int)
    parser.add_argument("--project_name", default='NuImages_swin_base_Seg', type=str)

    args = parser.parse_args()
    cfg_ake150 = Config.fromfile(args.config)

    cfg_base = CfgNode.load_yaml_with_base(args.config, allow_unsafe=True)    
    cfg_base.update(cfg_ake150.__dict__.items())

    cfg = cfg_base
    for k, v in args.__dict__.items():
        cfg[k] = v

    cfg = Config(cfg)

    cfg.ngpus = torch.cuda.device_count()
    if torch.cuda.device_count() > 1:
        cfg.local_rank = torch.distributed.get_rank()
        torch.cuda.set_device(cfg.local_rank)
    return cfg


def train_ade200k():
    cfg = get_args()
    dataset_train = ADE200kDataset(cfg.DATASETS.TRAIN, cfg, dynamic_batchHW=True)
    if cfg.ngpus > 1:
        train_sampler = torch.utils.data.distributed.DistributedSampler(dataset_train, rank=cfg.local_rank)
    else:
        train_sampler = None                            
    loader_train = torch.utils.data.DataLoader(
        dataset_train,
        batch_size=cfg.TRAIN.BATCH_SIZE,
        shuffle=False if train_sampler is not None else True,  
        collate_fn=dataset_train.collate_fn,
        num_workers=cfg.TRAIN.WORKERS,
        drop_last=True,
        pin_memory=True,
        sampler=train_sampler)

    dataset_eval = ADE200kDataset(cfg.DATASETS.VALID, cfg)
    loader_eval = torch.utils.data.DataLoader(
        dataset_eval,
        batch_size=1,
        shuffle=False,  
        collate_fn=dataset_eval.collate_fn,
        num_workers=cfg.TRAIN.WORKERS)

    seg_model = MaskFormer(cfg)
    seg_model.train(train_sampler, loader_train, loader_eval, cfg.TRAIN.EPOCH)

def train_nuimages():
    cfg = get_args()
    dataset_train = NuImagesDataset(cfg.DATASETS.ROOT_DIR, cfg, version='v1.0-train') # v1.0-mini or v1.0-train
    dataset_eval = NuImagesDataset(cfg.DATASETS.ROOT_DIR, cfg, version='v1.0-val') # v1.0-mini or v1.0-val

    if cfg.ngpus > 1:
        train_sampler = torch.utils.data.distributed.DistributedSampler(dataset_train, rank=cfg.local_rank)
        eval_sampler = torch.utils.data.distributed.DistributedSampler(dataset_eval, rank=cfg.local_rank)
    else:
        train_sampler = None     
        eval_sampler = None

    loader_train = torch.utils.data.DataLoader(
        dataset_train,
        batch_size=cfg.TRAIN.BATCH_SIZE,
        shuffle=False if train_sampler is not None else True,  
        collate_fn=dataset_train.collate_fn,
        num_workers=cfg.TRAIN.WORKERS,
        drop_last=True,
        pin_memory=True,
        sampler=train_sampler)
    
    loader_eval = torch.utils.data.DataLoader(
        dataset_eval,
        batch_size=1,
        shuffle=False if eval_sampler is not None else True,  
        collate_fn=dataset_eval.collate_fn,
        num_workers=cfg.TRAIN.WORKERS,
        drop_last=False,
        pin_memory=True,
        sampler=eval_sampler)

    seg_model = MaskFormer(cfg)
    seg_model.train(train_sampler, loader_train, loader_eval, cfg.TRAIN.EPOCH)

def segmentation_test():
    cfg = get_args()
    segmentation_handler = Segmentation(cfg)
    segmentation_handler.forward()
    

if __name__ == '__main__':
    train_nuimages()
    # segmentation_test()