event-ahu / coesot Goto Github PK

Hello, thank you very much for your open source! Can you please tell me that in the processing of COESOT dataset, running python coesot_choose_videos.py, there exists too many empty, how do you deal with it?

in CEUtrack/lib/train/dataset/ceosot.py line32
seq_ids = pandas.read_csv(file_path, header=None, dtype=np.int64).squeeze("columns").values.tolist()
During the reading process, the following issues may arise:
invalid literal for int() with base 10: 'dvSave-2021_09_01_06_59_10'

May I ask what the reasons are and how to solve them?

Thank you

Could you provide the raw results of CEUTrack on FE108 and VisEvent?

Hello, I would like to ask some questions:
Checkpoints path: /COESOT-main/CEUTrack/output/checkpoints/train/ceutrack/CEUTrack_ep0049.pth.tar
Training:
export CUDA_VISIBLE_DEVICES=1
python tracking/train.py --script ceutrack --config ceutrack_coesot --save_dir . /output --mode single --nproc_per_node 1 --use_wandb 0
Display: No matching checkpoint file found.

My path is set correctly, but why does running it say chekpoints not found?

There no link to download the model weights, could you kindly provide it?

Thanks for your work. Since your dataset “COESOT” is not downloaded yet, I would like to run the code on "Visevent" first.
In "lib.train.dataset.visevent.py", within the "_get_sequence_list()" function, there is an operation to open the "list.txt" file. I would like to know what content is associated with the Visevent dataset in "list.txt".

Hi, thanks for your work! This is my first time running the event-based tracking code. I have a simple question and hope to get your help.
I want to evaluate the FE108 dataset with this code. The current directory of the dataset is as shown below (only show the directory under /FE108/train/airplane/).

If we run ' python tracking/test.py ceutrack ceutrack_coesot --dataset fe108' directly, the following error will be displayed.

Then, I run 'python /CEUTrack/scripts/event_to_voxel_FE108.py'. After run this command, each sequence will be processed into 3 folders: voxel, aps, dvs. But only the dvs folder has data (some *.png), the other two folders are empty. Finally, I still can't evaluate the FE108 dataset.

Hi! Thank you for creating the coesot dataset.
When I used this dataset, I wonder if there is any csv.file in this dataset, because when I following the README instruction, the pycharm report error in coesot.py 32 row.
I found train.txt you provided could not be read_csv in np.int64, so I want to know how to read this file.

Thanks for your work. As mentioned in the paper, 205 training sequences and 171 testing videos in VisEvent dataset are selected to evaluate the tracker’s performance. Could you offer the used video's name? And I also cannot find the raw results and trained models of VisEvent Dataset and FE108 Dataset.

    if split is not None:
        if seq_ids is not None:
            raise ValueError('Cannot set both split_name and seq_ids.')
        if split == 'train':
            file_path = os.path.join(self.root, 'train.txt')
        elif split == 'val':
            file_path = os.path.join(self.root, 'val.txt')
        else:
            raise ValueError('Unknown split name')
        seq_ids = pandas.read_csv(file_path, header=None, dtype=np.int64).squeeze("columns").values.tolist()

def _get_sequence_list(self):
    with open(os.path.join(self.root, 'list.txt')) as f:
        dir_list = list(csv.reader(f))
    dir_list = [dir_name[0] for dir_name in dir_list]
    return dir_list

There're no such files "train.txt" "val.txt" "list.txt"，where can i get them

As you say in the paper, the number of the testing subset is 527, however, when I download the test dataset, I get the number of 528 (the number of the training is right 827), I guess if there is any mistake in the link of the baidu yun.

Best wishes

Hi, in event_to_voxel_coesot.py you set

if foldName in ['dvSave-2022_01_26_20_01_46', 'dvSave-2022_03_21_11_09_50',
'dvSave-2022_03_21_16_11_40', 'dvSave-2022_03_21_11_12_27',
'dvSave-2022_03_21_11_19_37', 'dvSave-2022_02_25_13_21_41',
'dvSave-2022_02_25_13_31_18', 'dvSave-2022_03_21_09_05_49'].
print("==>> pass it: ", foldName)
continue

but there are sequences in train.txt and test.txt that load some of this data, and in training, the voxel file of the data sequences is needed, so should it be commented out in event_to_voxel_coesot.py?

I just downlaod you provided tracking results and plot SR, PR, NPR figures. But I find SR figure is same with yours. But PR and NPR values are different?