zyayoung / iter-deformable-detr Goto Github PK

Dear author,

Could you release the pretrained SwinL again?, looks like the uploaded file is broken, I couldn't open it using any zip editor, and as a result, the code isn't working with this backbone ...

Hi, thank you for your open codes!
Could you open your detail configs on Citypersons?
Thanks!

when running your code on kaggle server this error appears to me

File "/kaggle/input/iter302/Iter-Deformable-DETR-main/detr/deformable_transformer.py", line 395, in forward
pred_boxes = self.bbox_embedlid + inverse_sigmoid(reference_points)
TypeError: 'NoneType' object is not subscriptable

In your code, you initialize bbox_embed as:
self.bbox_embed = None
self.class_embed = None

then in the forward pass to try to subscripe it.

when running your code on my own laptop with RTX2060 GPU, this error appears

assert (input_spatial_shapes[:, 0] * input_spatial_shapes[:, 1]).sum() == Len_in

RuntimeError: CUDA error: an illegal memory access was encountered
CUDA kernel errors might be asynchronously reported at some other API call,so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1.

it doesn't work for me

I use the newest code, but when I evaluate it directly, I got the error:

  0%|                                                                                                                                                                                      | 0/4370 [00:00<?, ?it/s]loading annotations into memory...
Done (t=0.40s)
creating index...
index created!
Process Process-1:
Traceback (most recent call last):
  File "/home/sipl/anaconda3/envs/e2edet_new/lib/python3.8/multiprocessing/process.py", line 315, in _bootstrap
    self.run()
  File "/home/sipl/anaconda3/envs/e2edet_new/lib/python3.8/multiprocessing/process.py", line 108, in run
    self._target(*self._args, **self._kwargs)
  File "testing.py", line 166, in processor
    checkpoint = torch.load(model_file, map_location='cpu')
  File "/home/sipl/anaconda3/envs/e2edet_new/lib/python3.8/site-packages/torch/serialization.py", line 587, in load
    with _open_zipfile_reader(opened_file) as opened_zipfile:
  File "/home/sipl/anaconda3/envs/e2edet_new/lib/python3.8/site-packages/torch/serialization.py", line 242, in __init__
    super(_open_zipfile_reader, self).__init__(torch._C.PyTorchFileReader(name_or_buffer))
RuntimeError: [enforce fail at inline_container.cc:145] . PytorchStreamReader failed reading zip archive: failed finding central directory

Could you tell me how to solve it?

Thank you!!

Hello, thank you very much for your work. I first trained for 2 epochs, then started the test, and the result reported an error:

loading annotations into memory...
Done (t=0.32s)
creating index...
index created!
saved                                                                                                                                                                                                              
Traceback (most recent call last):
  File "demo_opt.py", line 472, in <module>
    eval_results = _evaluate_predictions_on_crowdhuman(gt_path, fpath)
  File "demo_opt.py", line 448, in _evaluate_predictions_on_crowdhuman
    database = Database(gt_path, dt_path, target_key, None, mode)
  File "demo_opt.py", line 312, in __init__
    self.loadData(dtpath, body_key, head_key, if_gt=False)
  File "demo_opt.py", line 331, in loadData
    self.images[record["ID"]].clip_all_boader()
  File "demo_opt.py", line 195, in clip_all_boader
    assert self._width is not None and self._height is not None
AssertionError

I followed the instructions completely, but why did it report an error?

Thank you so much for your incredible works,

I'm trying to evaluate your code using this command: bash exps/aps_test.sh 49 and encountering this issue, the height and width values don't exist in the "record" loaded by the output of testing.py, in specific, the epoch049.human file. As a result, these lines of code will return None:
if "width" in record and self._width is None: self._width = record["width"]
if "height" in record and self._height is None: self._height = record["height"]

And in the clip_all_boarder will face this error:
assert self._width is not None and self._height is not None

Do you have any idea to overcome this issue?

Evaluating the base model with the pretrained weights, I got this error.
I am running the code using windows anaconda environment on RTX 2060 GPU.
Any help?

When I try to eval with the pretained model.
I miss this error. I have the cofig right. Asking for your help.

Retraning your model on my RTX2060 GPU with cuda 10.2 version, I got this error

any help.......?!

The format of crowdhuman data is known as the odgt format.
Are you using the crowdhuman data format converted to coco format?

and Do I have to input both coco format json(train,eval) and odgt format json(train,eval) into config.py?
If so, how do I convert odgt to other custom data that is not in odgt format?

Hi,

Do you have any plan to release the inference code supporting custom images rather than benchmarking with predefined datasets?