scuwyh2000 / randbox Goto Github PK

Dear authors,

Your paper overlooks several significant works in the Open World Object Detection (OWOD) field. Specifically, the following papers from this year's CVPR conference:

1. "PROB: Probabilistic Objectness for Open World Object Detection"
2. "Annealing-Based Label-Transfer Learning for Open World Object Detection"
3. "CAT: LoCalization and IdentificAtion Cascade Detection Transformer for Open-World Object Detection"

As well as other significant works published in 2022, such as "OPCL" (ICIP2022) and "2B-OCD" (HCMA2022).

It is essential to acknowledge and include existing OWOD work to ensure the accuracy and completeness of your research. Failure to do so undermines the credibility of your work and disregards the contributions of others in the field.

Sincerely,
Orr Zohar

How long does it take for a 512 × 512 graph to be output in real time? Can it be converted to TRT?

How do you use your own images or videos for inference detection? Is there a script?

@scuwyh2000 你好
1.推理过程中报错:FileNotFoundError: [Errno 2] No such file or directory: 'datasets/test.txt',

Hello author, I conducted experiments according to your file configuration (completely consistent settings) and found that the experimental results are different from the paper results.

Task t1:
AOSE is about 1500 higher
K-map is 4 points lower
UR is 3 points lower
What causes this?

thanks for your work, however, when I run your code, I found some problem:

Traceback (most recent call last):
File "f:\documents_lry\detectron2\detectron2\engine\train_loop.py", line 155, in train
self.run_step()
File "f:\documents_lry\detectron2\detectron2\engine\defaults.py", line 496, in run_step
self._trainer.run_step()
File "f:\documents_lry\detectron2\detectron2\engine\train_loop.py", line 310, in run_step
loss_dict = self.model(data)
File "D:\Anaconda3\envs\detectron\lib\site-packages\torch\nn\modules\module.py", line 1102, in _call_impl
return forward_call(*input, **kwargs)
File "F:\Documents_lry\RandBox\randbox\detector.py", line 296, in forward
targets, x_boxes = self.prepare_targets(gt_instances)
File "F:\Documents_lry\RandBox\randbox\detector.py", line 350, in prepare_targets
d_boxes = self.prepare_concat(gt_boxes)
File "F:\Documents_lry\RandBox\randbox\detector.py", line 330, in prepare_concat
x = (x_start * 2. - 1.) * self.scale
NameError: name 'x_start' is not defined

there are not 'x_start' in function "self.prepare_concat(gt_boxes)", I can only found 'x_start' in function "model_predictions", could you please fix this problem?

请问t2和t2_ft的区别是什么

I have downloaded the Coco and VOC datasets. How can I divide these datasets into t1, t2, t2_ft..........

Hello. First of all, thank you for your work and for sharing it!

I am implementing the evaluation process of OWOD (defined in PascalVOCDetectionEvaluator.evaluate). I have encountered several problems to correctly apply the benchmark to my own work. Here are them:

• For evaluating every task and every metric (even WI), I assume you are always performing evaluation against all_task_test.txt. Is that right?
• For the Wilderness Impact, they get reported the WI values for different recalls. Which one do you chose?
• Are instances categorized as "difficult" used in for the metrics?
• The reported number of test instances of T1 in the below image of your paper refer only to the KNOWN classes or are accounting for both KNOWN and UNKNOWN objects?
  

Hi, thanks for your great work!

I am confused about the data preparation. According to your get_started, the data are from MSCOCO. However, in the ./split files, there are some image ids that seem not from COCO, such as 001299 or 2008_008613.

May I ask where are they from?

I dont find such a script

Thank author for your great work. I am wondering how to draw Figure 2 in the paper as shown blow:

I have the following questions about the implementation of Figure 2.

1. All the proposals have been shown in the Figure or only part has been shown.
2. The proposals are before or after NMS?

I hope the authors can give me some advice for representing this figure. Thanks a lot!

In the paper, you state that you evaluate models trained on Task 1 on the 4,810 validation images of LVIS v1.0, however, there are 19809 images in the validation set of LVIS v1.0. Will you release the images list used for transfer learning?

Your paper described the evaluation protocol as:

the final predictions are the bboxes whose probability of belonging to a class in K ∪ {“unknown′′} are at least 0.1.

But your code uses a different threshold:

Hi, I got a error when run run_eval.sh, can you give me some help?

Traceback (most recent call last):
  File "train_net.py", line 376, in <module>
    launch(
  File "/mnt/home/wls/anaconda3/envs/randbox/lib/python3.8/site-packages/detectron2/engine/launch.py", line 67, in launch
    mp.spawn(
  File "/mnt/home/wls/anaconda3/envs/randbox/lib/python3.8/site-packages/torch/multiprocessing/spawn.py", line 230, in spawn
    return start_processes(fn, args, nprocs, join, daemon, start_method='spawn')
  File "/mnt/home/wls/anaconda3/envs/randbox/lib/python3.8/site-packages/torch/multiprocessing/spawn.py", line 188, in start_processes
    while not context.join():
  File "/mnt/home/wls/anaconda3/envs/randbox/lib/python3.8/site-packages/torch/multiprocessing/spawn.py", line 150, in join
    raise ProcessRaisedException(msg, error_index, failed_process.pid)
torch.multiprocessing.spawn.ProcessRaisedException: 

-- Process 0 terminated with the following error:
Traceback (most recent call last):
  File "/mnt/home/wls/anaconda3/envs/randbox/lib/python3.8/site-packages/torch/multiprocessing/spawn.py", line 59, in _wrap
    fn(i, *args)
  File "/mnt/home/wls/anaconda3/envs/randbox/lib/python3.8/site-packages/detectron2/engine/launch.py", line 126, in _distributed_worker
    main_func(*args)
  File "/mnt/home/wls/RandBox/train_net.py", line 359, in main
    res = Trainer.ema_test(cfg, model)
  File "/mnt/home/wls/RandBox/train_net.py", line 261, in ema_test
    results = cls.test(cfg, model, evaluators=evaluators)
  File "/mnt/home/wls/anaconda3/envs/randbox/lib/python3.8/site-packages/detectron2/engine/defaults.py", line 608, in test
    results_i = inference_on_dataset(model, data_loader, evaluator)
  File "/mnt/home/wls/anaconda3/envs/randbox/lib/python3.8/site-packages/detectron2/evaluation/evaluator.py", line 204, in inference_on_dataset
    results = evaluator.evaluate()
  File "/mnt/home/wls/RandBox/randbox/pascal_voc_evaluation.py", line 189, in evaluate
    rec, prec, ap, unk_det_as_known, num_unk, tp_plus_fp_closed_set, fp_open_set = voc_eval(
  File "/mnt/home/wls/RandBox/randbox/pascal_voc_evaluation.py", line 456, in voc_eval
    R = class_recs[str(mapping[int(image_ids[d])])]
KeyError: '009122'

你好，请问代码更新完了吗

@scuwyh2000
Hi, is the test.json under each task the same and if not the test.json under each task is generated based on which file? Currently there is only all_task_test.txt, how to generate the json needed for each task?

Thank you for your excellent work.
In your supplementary material, you described the pre-defined inference bboxes.
Are the pre-defined bboxes in the code and described in the paper the same?

Are the data sets used in the paper the same as those used in ORE (Towards Open World Object Detection)? Can I directly use the link to the dataset provided in ORE?