question about evaluate about bbavectors-oriented-object-detection HOT 16 OPEN

It looks like your ResultMerge_multi_process.py has some issues. You may need to debug the code. For example, did you generate the result_dota folder before you call dsets.merge_crop_image_results(result_path, merge_path) in eval.py? You can use online server https://captain-whu.github.io/DOTA/evaluation.html to calculate AP dota. The hrsc dataset is calculated by ap = dsets.dec_evaluation(result_path) in eval.py.

from bbavectors-oriented-object-detection.

Thank you very much.Before I call dsets.merge_crop_image_results(result_path, merge_path) in eval.py,It has genrate the result_dota folder

from bbavectors-oriented-object-detection.

Does the result_dota contain the right files? like Task1_baseball-diamond.txt files contain

P0045__0.5__0___500 0.683183252811 446.0 444.2 465.2 560.3 348.0 578.1 328.8 461.9
P0045__0.5__0___500 0.636462032795 407.4 252.5 429.9 360.8 315.2 383.2 292.8 274.9
P0045__0.5__0___500 0.531519293785 592.8 219.0 615.2 325.4 516.1 345.5 493.7 239.1
P0045__0.5__0___500 0.160389885306 593.1 411.4 614.1 517.1 547.6 532.2 526.6 426.5
P0045__0.5__0___803 0.659127235413 445.5 139.0 467.2 246.9 355.6 267.6 333.9 159.7
P0045__0.5__0___803 0.542766690254 348.1 382.6 370.6 490.0 255.8 513.1 233.3 405.8

from bbavectors-oriented-object-detection.

yes ,it have

from bbavectors-oriented-object-detection.

Sorry I have no idea. You may need to debug to see what makes the index out of range.

from bbavectors-oriented-object-detection.

When I evaluate the model by phase eva , a error is occured.Could you help me to solve it ?
And I want to ask you how to calculate the ap or map?
Thank you,Best wish

我也遇到这个问题了，请问您解决了吗？

from bbavectors-oriented-object-detection.

By debugging I found the error causes in DOTA_devkit.ResultMerge_multi_process:line 199-202
but to utilize debugging better, I recommend using ResultMerge instead of ResultMerge_multi_process in datasets.dataset_dota:line 5
In my case I convert HRSC2016 format to dota format, and results from HRSC2016 as following:

100000849 0.318158239126 53.9 147.5 100.6 211.7 29.2 238.1 -17.4 173.9
100001347 0.189944535494 6.5 336.8 38.7 353.3 18.7 414.5 -13.6 398.0

But in dota, it will use filename to get metadata, like ratio=0.5 x=0 y=500 from P0045__0.5__0___803 0.542766690254 348.1 382.6 370.6 490.0 255.8 513.1 233.3 405.8
so when module re find nothing from my file name, it will cause list index out of range by x_y_2 = re.findall(r'\d+', x_y[0])...

from bbavectors-oriented-object-detection.

By debugging I found the error causes in DOTA_devkit.ResultMerge_multi_process:line 199-202
but to utilize debugging better, I recommend using ResultMerge instead of ResultMerge_multi_process in datasets.dataset_dota:line 5
In my case I convert HRSC2016 format to dota format, and results from HRSC2016 as following:

100000849 0.318158239126 53.9 147.5 100.6 211.7 29.2 238.1 -17.4 173.9
100001347 0.189944535494 6.5 336.8 38.7 353.3 18.7 414.5 -13.6 398.0

But in dota, it will use filename to get metadata, like ratio=0.5 x=0 y=500 from P0045__0.5__0___803 0.542766690254 348.1 382.6 370.6 490.0 255.8 513.1 233.3 405.8
so when module re find nothing from my file name, it will cause list index out of range by x_y_2 = re.findall(r'\d+', x_y[0])...

I see. So it looks like if the data is not arranged in the right way (e.g. merge format), it will raise exceptions. Thanks @igo312

from bbavectors-oriented-object-detection.

I want to ask if the pictures of test set also be split

from bbavectors-oriented-object-detection.

Does the result_dota contain the right files? like Task1_baseball-diamond.txt files contain

P0045__0.5__0___500 0.683183252811 446.0 444.2 465.2 560.3 348.0 578.1 328.8 461.9
P0045__0.5__0___500 0.636462032795 407.4 252.5 429.9 360.8 315.2 383.2 292.8 274.9
P0045__0.5__0___500 0.531519293785 592.8 219.0 615.2 325.4 516.1 345.5 493.7 239.1
P0045__0.5__0___500 0.160389885306 593.1 411.4 614.1 517.1 547.6 532.2 526.6 426.5
P0045__0.5__0___803 0.659127235413 445.5 139.0 467.2 246.9 355.6 267.6 333.9 159.7
P0045__0.5__0___803 0.542766690254 348.1 382.6 370.6 490.0 255.8 513.1 233.3 405.8

When I execute the eval code, the result_dota file is empty. Is there a problem with my storage path?

from bbavectors-oriented-object-detection.

I want to ask if the pictures of test set also be split

Yes, the dota test set is also split.

from bbavectors-oriented-object-detection.

Does the result_dota contain the right files? like Task1_baseball-diamond.txt files contain

P0045__0.5__0___500 0.683183252811 446.0 444.2 465.2 560.3 348.0 578.1 328.8 461.9
P0045__0.5__0___500 0.636462032795 407.4 252.5 429.9 360.8 315.2 383.2 292.8 274.9
P0045__0.5__0___500 0.531519293785 592.8 219.0 615.2 325.4 516.1 345.5 493.7 239.1
P0045__0.5__0___500 0.160389885306 593.1 411.4 614.1 517.1 547.6 532.2 526.6 426.5
P0045__0.5__0___803 0.659127235413 445.5 139.0 467.2 246.9 355.6 267.6 333.9 159.7
P0045__0.5__0___803 0.542766690254 348.1 382.6 370.6 490.0 255.8 513.1 233.3 405.8

When I execute the eval code, the result_dota file is empty. Is there a problem with my storage path?

If you are testing dota, make sure your test set is split as the trainval set.

from bbavectors-oriented-object-detection.

I want to ask if the pictures of test set also be split

Yes, the dota test set is also split.

您好，我用了ROI Transformer的源码分割了DOTA数据集，但是我没有得到您论文里说的那么多张图片。我最后复现的结果map只有69.7%左右，不知道和图片分割有没有关系。

from bbavectors-oriented-object-detection.

I want to ask if the pictures of test set also be split

Yes, the dota test set is also split.

您好，我用了ROI Transformer的源码分割了DOTA数据集，但是我没有得到您论文里说的那么多张图片。我最后复现的结果map只有69.7%左右，不知道和图片分割有没有关系。

你好，请问你是分割成608*608大小的图片吗

from bbavectors-oriented-object-detection.

I want to ask if the pictures of test set also be split

Yes, the dota test set is also split.

您好，我用了ROI Transformer的源码分割了DOTA数据集，但是我没有得到您论文里说的那么多张图片。我最后复现的结果map只有69.7%左右，不知道和图片分割有没有关系。

你好，请问你是分割成608*608大小的图片吗

可能不是，我没有改代码直接分割的，作者的意思是要分割成608*608大小的图片吗

from bbavectors-oriented-object-detection.

I want to ask if the pictures of test set also be split

Yes, the dota test set is also split.

您好，我用了ROI Transformer的源码分割了DOTA数据集，但是我没有得到您论文里说的那么多张图片。我最后复现的结果map只有69.7%左右，不知道和图片分割有没有关系。

In particular, the images are cropped into 600×600 patches with a stride of 100. The input images have two scales 0.5 and 1.

from bbavectors-oriented-object-detection.