Mmdetection3d (2) -- visualization of results and logs

１． Result visualization ：
Installed docker Version of mmdetection3d There are some problems in visualization ,open3d It is always difficult to open the window , But the installed version of the source code can realize visualization . Therefore, the source code installation method is adopted . The source code installation is also relatively simple , If the environment is ok , It's a smooth process .
（１） Source code installation steps
Installation process method ：https://github.com/open-mmlab/mmdetection3d/blob/master/docs/getting_started.md
There are no pits during installation , Just follow the steps step by step .

After installation and operation, something like :

KeyError: 'SparseConv2d is already registered in conv layer'

The error of , The approximate rate is MMCV Version too high , Appropriate reduction mmcv edition , Just reinstall .

python_open3d_Error: The DISPLAY environment variable is missing
https://blog.csdn.net/shyjhyp11/article/details/113250325

（2） Use the installed version of the source code for visualization
First, configure config In file test_cfg Partial iou_thre as well as score_thr
a. Online visualization
After running, each frame will be visualized , And after visualization, the original point cloud and prediction results will be .obj Save the file to show-dir Under the table of contents . Every time you want to switch to the next frame, you will open3d window × fall . However, this method can only show the prediction results ,GT Not shown together .

eg:
python tools/test.py configs/3dssd/3dssd_kitti-3d-car.py data/mmdetection/pretrain_weight/3dssd_kitti-3d-car_20210324_122002-07e9a19b.pth --show --show-dir ./data/zjh_data/mmdetection/show_result

b. Offline visualization , You can talk to GT Show together
Want and GT Show together , The files to be generated also include gt.obj. Use the following commands to get gt.obj

python tools/test.py ${
    CONFIG_FILE} ${
    CKPT_PATH} --eval 'mAP' --eval-options 'show=True' 'out_dir=${SHOW_DIR}'

But in actual operation , My source code installed mmdet3d The following error occurred .
This problem has not been solved for the time being , Take a note of it . I haven't changed this part of the source code , I think there is something wrong with the environment configuration .

File "/mnt/data/zjh_data/mmdet3d_install/mmdetection3d/mmdet3d/core/evaluation/kitti_utils/eval.py", line 365, in calculate_iou_partly
    overlap_part = image_box_overlap(gt_boxes, dt_boxes)
TypeError: expected dtype object, got 'numpy.dtype[float64]'

Curve saving method ：
use docker Mr. version becomes a visual file , Then use the visualization of the source code version . however docker Version because open3d The problem of , It makes it impossible to visualize, which leads to the call open3d When you're in trouble , Makes it impossible to generate visualization files .
resolvent ： Open Directory mmdet3d/core/visualization/show_result.py file , hold show_result Function if show　 call open3d All parts are annotated , Not because of open3d You cannot use the influence visualization file to generate .
Run the following command to generate the model .pkl Files and visualization files 000023_gt.obj ,000023_points.obj ,000023_pred.obj.

python tools/test.py ${
    CONFIG_FILE} ${
    CKPT_PATH} --out ./data/result_output/out_dir/3dssd.pkl --eval 'mAP' --eval-options 'show=True' 'out_dir=./data/result_output/show_result'

After generation, use tool Offline visualization script in ：

python tools/misc/visualize_results.py ${
    CONFIG_FILE} --result ~/zjh_data/result_output/out_dir/3dssd.pkl --show-dir ~/zjh_data/result_output/show_result/

2.log Visualization of error curve in file

 python tools/analysis_tools/analyze_logs.py plot_curve ~/zjh_data/mmdetection/work_dir623/20210624_004729.log.json --keys loss --legend loss --out loss.pdf

3. test Test and submit to kitti Online evaluation on
   （1） modify config In file

ann_file=data_root + 'kitti_infos_test.pkl',
        split='testing',

data = dict(
    samples_per_gpu=6,
    workers_per_gpu=4,
    train=dict(
        type='RepeatDataset',
        times=2,
        dataset=dict(
            type=dataset_type,
            data_root=data_root,
            ann_file=data_root + 'kitti_infos_train.pkl',
            split='training',
            pts_prefix='velodyne_reduced',
            pipeline=train_pipeline,
            modality=input_modality,
            classes=class_names,
            test_mode=False,
            # we use box_type_3d='LiDAR' in kitti and nuscenes dataset
            # and box_type_3d='Depth' in sunrgbd and scannet dataset.
            box_type_3d='LiDAR')),
    val=dict(
        type=dataset_type,
        data_root=data_root,
        ann_file=data_root + 'kitti_infos_val.pkl',
        split='training',
        pts_prefix='velodyne_reduced',
        pipeline=test_pipeline,
        modality=input_modality,
        classes=class_names,
        test_mode=True,
        box_type_3d='LiDAR'),
    test=dict(
        type=dataset_type,
        data_root=data_root,
        #  It has been changed here , When training, change to val
        #### The next two lines are revised to test data 
        ann_file=data_root + 'kitti_infos_test.pkl',
        split='testing',
        pts_prefix='velodyne_reduced',
        pipeline=test_pipeline,
        modality=input_modality,
        classes=class_names,
        test_mode=True,
        box_type_3d='LiDAR'))

（2） Use script files to test