Hi， thanks for your Excellent work! I have a question about the wireless table of icdar2019. In your paper, you said 'For lacking wireless table in WTW, we finetune the Cycle-CenterNet on the ICDAR2019 trainingset and test on ICDAR 2019 Track B2 dataset', But, icdar2019 TrackB.2 ‘Morden Dataset’ only has 100 test sets. TrackB,1 only has 'Historical Datasets'. Which wireless dataset did you use in your paper?

我以xml中各单元格的start_row,start_col,end_row,end_col为标准，生成了html标注，发现很多图中单元格的跨行跨列有问题，本应没有跨行跨列的表格被标注成了跨多行/多列，确认到该现象一旦出现，在一系列相似的图片中都会出现。 例如：mit_google_image_search-10918758-d6cc32fbb935608d01f71d2c3daa7ebd6634aabb

这些标注问题貌似对邻接关系与TEDS的评测都有很大影响。

Hi Wangwen,

Excellent work with WTW dataset. I wanted to download data, it just doesn't downloads, even after signing in and agreeing to the consent. Is there any other way I can download the data. Thanks in advance.

Sincerely, Om Rastogi

Hello, I have some questions about the results of the test set in the paper. In the paper, you divide the test set into 7 different sub-categories, which contain 'Irregular tables'. However, in Table 4, there is no result of "Irregular tables", instead there is a result of "Simple tables" which has not been mentioned before. How do you understand this?

在使用centernet的代码复现其physical structure中的P,R,F指标时，发现自己的结果为 在iou=0.9时检测结果为0 并可视化了一些结果，发现其大概是正常的 所以对您iou=0.9的指标有些疑惑 我使用了3090x4来复现您的结果其opt如下

==> torch version: 1.7.0+cu110 ==> cudnn version: 8004 ==> Cmd: ['src/main.py', 'ctdet', '--exp_id', 'wtw_lr_1024', '--dataset', 'wtw', '--batch_size', '32', '--master_batch', '9', '--lr', '1.25e-4', '--gpus', '0,1,2,3', '--num_epochs', '150', '--lr_step', '90,120', '--input_res', '1024'] ==> Opt: K: 100 aggr_weight: 0.0 agnostic_ex: False arch: dla_34 aug_ddd: 0.5 aug_rot: 0 batch_size: 32 cat_spec_wh: False center_thresh: 0.1 chunk_sizes: [9, 8, 8, 7] data_dir: /home/mdisk2/bianzhe/CenterNet/src/lib/../../data dataset: wtw debug: 0 debug_dir: /home/mdisk2/bianzhe/CenterNet/src/lib/../../exp/ctdet/wtw_lr_1024/debug debugger_theme: white demo: dense_hp: False dense_wh: False dep_weight: 1 dim_weight: 1 down_ratio: 4 eval_oracle_dep: False eval_oracle_hm: False eval_oracle_hmhp: False eval_oracle_hp_offset: False eval_oracle_kps: False eval_oracle_offset: False eval_oracle_wh: False exp_dir: /home/mdisk2/bianzhe/CenterNet/src/lib/../../exp/ctdet exp_id: wtw_lr_1024 fix_res: True flip: 0.5 flip_test: False gpus: [0, 1, 2, 3] gpus_str: 0,1,2,3 head_conv: 256 heads: {'hm': 1, 'wh': 2, 'reg': 2} hide_data_time: False hm_hp: True hm_hp_weight: 1 hm_weight: 1 hp_weight: 1 input_h: 1024 input_res: 1024 input_w: 1024 keep_res: False kitti_split: 3dop load_model: lr: 0.000125 lr_step: [90, 120] master_batch_size: 9 mean: [[[0.7733555 0.78804266 0.7961663 ]]] metric: loss mse_loss: False nms: False no_color_aug: False norm_wh: False not_cuda_benchmark: False not_hm_hp: False not_prefetch_test: False not_rand_crop: False not_reg_bbox: False not_reg_hp_offset: False not_reg_offset: False num_classes: 1 num_epochs: 150 num_iters: -1 num_stacks: 1 num_workers: 4 off_weight: 1 output_h: 256 output_res: 256 output_w: 256 pad: 31 peak_thresh: 0.2 print_iter: 0 rect_mask: False reg_bbox: True reg_hp_offset: True reg_loss: l1 reg_offset: True resume: False root_dir: /home/mdisk2/bianzhe/CenterNet/src/lib/../.. rot_weight: 1 rotate: 0 save_all: False save_dir: /home/mdisk2/bianzhe/CenterNet/src/lib/../../exp/ctdet/wtw_lr_1024 scale: 0.4 scores_thresh: 0.1 seed: 317 shift: 0.1 std: [[[0.28359458 0.26819503 0.26328853]]] task: ctdet test: False test_scales: [1.0] trainval: False val_intervals: 5 vis_thresh: 0.3 wh_weight: 0.1

期待您的回复！

文中您提到的是采用ICDAR2013比赛指标（P, R, F1）进行计算，对此我有两个疑问： 1、该指标需要text content进行比较，而WTW数据集没有内容的标注，这里您是怎么计算指标的呢？ 2、该指标是不计算空单元格之间的匹配的，而WTW数据集同样没有标注单元格是否为空，请问这里是怎么处理的？

（个人猜测：您是否使用的是ICDAR2019的比赛指标？是否将所有单元格均当做非空单元格？）

Thank you for your excellent paper and the dataset. Could you please share the dataset annotation tool? cause I found it is hard to label the table with normal annotation tools since the cells sharing common vertices.

All  the experiments are performed on a
workstation with 8 NVIDIA GTX 1080Ti GPUs. During
the training, we set the batch size to 32 per GPU in parallel

In the paper, i found the batchSize is 32 per GPU，Is there an error here？The experiment I did maximum batch_size is 4 per GPU

When I do my sub-category experiment in the sub-categories in WTW, because there may be more than one table in a single image, I tried to assemble the sub-cate result in your paper to find out the TEDS are calculated per image or per table, but both ways not matched the TEDS score of the whole test set.

我在做子类别实验的时候，因为一幅图里可能有几张表，所以我想知道你们论文的子类别实验到底是 TEDS/每图 还是 TEDS/每表，但是我用子类别TEDS指标加起来算总指标的时候，两种方式都对不上，所以我想知道你们是用什么规则计算TEDS的，或者干脆就是我算错了。。。

我是这样算的： The formula below is my calculation method： PER IMAGE: from 'simple' to 'multicolor-grid' result = (94.293 + 90.6670 + .... + 66.7*749) /3611 = 76.17

PER TABLE: from 'simple' to 'multicolor-grid' result = (94.293 + 90.6676 + .... + 66.7*991) /4048 = 75.6

These reults not match any score in your paper What's wrong with that?Maybe I'm just stupid but please tell me the right method.