gather_models.py

# Copyright (c) OpenMMLab. All rights reserved.
"""Script to gather benchmarked models and prepare them for upload.

Usage:
python gather_models.py ${root_path} ${out_dir}

Example:
python gather_models.py \
work_dirs/pgd_r101_caffe_fpn_gn-head_3x4_4x_kitti-mono3d \
work_dirs/pgd_r101_caffe_fpn_gn-head_3x4_4x_kitti-mono3d

Note that before running the above command, rename the directory with the
config name if you did not use the default directory name, create
a corresponding directory 'pgd' under the above path and put the used config
into it.
"""

import argparse
import glob
import json
import mmcv
import shutil
import subprocess
import torch
from os import path as osp

# build schedule look-up table to automatically find the final model
SCHEDULES_LUT = {
    '_1x_': 12,
    '_2x_': 24,
    '_20e_': 20,
    '_3x_': 36,
    '_4x_': 48,
    '_24e_': 24,
    '_6x_': 73,
    '_50e_': 50,
    '_80e_': 80,
    '_100e_': 100,
    '_150e_': 150,
    '_200e_': 200,
    '_250e_': 250,
    '_400e_': 400
}

# TODO: add support for lyft dataset
RESULTS_LUT = {
    'coco': ['bbox_mAP', 'segm_mAP'],
    'nus': ['pts_bbox_NuScenes/NDS', 'NDS'],
    'kitti-3d-3class': ['KITTI/Overall_3D_moderate', 'Overall_3D_moderate'],
    'kitti-3d-car': ['KITTI/Car_3D_moderate_strict', 'Car_3D_moderate_strict'],
    'lyft': ['score'],
    'scannet_seg': ['miou'],
    's3dis_seg': ['miou'],
    'scannet': ['mAP_0.50'],
    'sunrgbd': ['mAP_0.50'],
    'kitti-mono3d': [
        'img_bbox/KITTI/Car_3D_AP40_moderate_strict',
        'Car_3D_AP40_moderate_strict'
    ],
    'nus-mono3d': ['img_bbox_NuScenes/NDS', 'NDS']
}


def get_model_dataset(log_json_path):
    for key in RESULTS_LUT:
        if log_json_path.find(key) != -1:
            return key


def process_checkpoint(in_file, out_file):
    checkpoint = torch.load(in_file, map_location='cpu')
    # remove optimizer for smaller file size
    if 'optimizer' in checkpoint:
        del checkpoint['optimizer']
    # if it is necessary to remove some sensitive data in checkpoint['meta'],
    # add the code here.
    torch.save(checkpoint, out_file)
    sha = subprocess.check_output(['sha256sum', out_file]).decode()
    final_file = out_file.rstrip('.pth') + '-{}.pth'.format(sha[:8])
    subprocess.Popen(['mv', out_file, final_file])
    return final_file


def get_final_epoch(config):
    if config.find('grid_rcnn') != -1 and config.find('2x') != -1:
        # grid_rcnn 2x trains 25 epochs
        return 25

    for schedule_name, epoch_num in SCHEDULES_LUT.items():
        if config.find(schedule_name) != -1:
            return epoch_num


def get_best_results(log_json_path):
    dataset = get_model_dataset(log_json_path)
    max_dict = dict()
    max_memory = 0
    with open(log_json_path, 'r') as f:
        for line in f.readlines():
            log_line = json.loads(line)
            if 'mode' not in log_line.keys():
                continue

            # record memory and find best results & epochs
            if log_line['mode'] == 'train' \
                    and max_memory <= log_line['memory']:
                max_memory = log_line['memory']

            elif log_line['mode'] == 'val':
                result_dict = {
                    key: log_line[key]
                    for key in RESULTS_LUT[dataset] if key in log_line
                }
                if len(max_dict) == 0:
                    max_dict = result_dict
                    max_dict['epoch'] = log_line['epoch']
                elif all(
                    [max_dict[key] <= result_dict[key]
                     for key in result_dict]):
                    max_dict.update(result_dict)
                    max_dict['epoch'] = log_line['epoch']

        max_dict['memory'] = max_memory
        return max_dict


def parse_args():
    parser = argparse.ArgumentParser(description='Gather benchmarked models')
    parser.add_argument(
        'root',
        type=str,
        help='root path of benchmarked models to be gathered')
    parser.add_argument(
        'out', type=str, help='output path of gathered models to be stored')

    args = parser.parse_args()
    return args


def main():
    args = parse_args()
    models_root = args.root
    models_out = args.out
    mmcv.mkdir_or_exist(models_out)

    # find all models in the root directory to be gathered
    raw_configs = list(mmcv.scandir('./configs', '.py', recursive=True))

    # filter configs that is not trained in the experiments dir
    used_configs = []
    for raw_config in raw_configs:
        if osp.exists(osp.join(models_root, raw_config)):
            used_configs.append(raw_config)
    print(f'Find {len(used_configs)} models to be gathered')

    # find final_ckpt and log file for trained each config
    # and parse the best performance
    model_infos = []
    for used_config in used_configs:
        # get logs
        log_json_path = glob.glob(osp.join(models_root, '*.log.json'))[0]
        log_txt_path = glob.glob(osp.join(models_root, '*.log'))[0]
        model_performance = get_best_results(log_json_path)
        final_epoch = model_performance['epoch']
        final_model = 'epoch_{}.pth'.format(final_epoch)
        model_path = osp.join(models_root, final_model)

        # skip if the model is still training
        if not osp.exists(model_path):
            print(f'Expected {model_path} does not exist!')
            continue

        if model_performance is None:
            print(f'Obtained no performance for model {used_config}')
            continue

        model_time = osp.split(log_txt_path)[-1].split('.')[0]
        model_infos.append(
            dict(
                config=used_config,
                results=model_performance,
                epochs=final_epoch,
                model_time=model_time,
                log_json_path=osp.split(log_json_path)[-1]))

    # publish model for each checkpoint
    publish_model_infos = []
    for model in model_infos:
        model_publish_dir = osp.join(models_out, model['config'].rstrip('.py'))
        mmcv.mkdir_or_exist(model_publish_dir)

        model_name = model['config'].split('/')[-1].rstrip(
            '.py') + '_' + model['model_time']
        publish_model_path = osp.join(model_publish_dir, model_name)
        trained_model_path = osp.join(models_root,
                                      'epoch_{}.pth'.format(model['epochs']))

        # convert model
        final_model_path = process_checkpoint(trained_model_path,
                                              publish_model_path)

        # copy log
        shutil.copy(
            osp.join(models_root, model['log_json_path']),
            osp.join(model_publish_dir, f'{model_name}.log.json'))
        shutil.copy(
            osp.join(models_root, model['log_json_path'].rstrip('.json')),
            osp.join(model_publish_dir, f'{model_name}.log'))

        # copy config to guarantee reproducibility
        config_path = model['config']
        config_path = osp.join(
            'configs',
            config_path) if 'configs' not in config_path else config_path
        target_cconfig_path = osp.split(config_path)[-1]
        shutil.copy(config_path,
                    osp.join(model_publish_dir, target_cconfig_path))

        model['model_path'] = final_model_path
        publish_model_infos.append(model)

    models = dict(models=publish_model_infos)
    print(f'Totally gathered {len(publish_model_infos)} models')
    mmcv.dump(models, osp.join(models_out, 'model_info.json'))


if __name__ == '__main__':
    main()