Open Flock1 opened 5 years ago
Hi @Flock1 go to yolo.py change the line if predicted_class != 'person' to if predicted_class != 'truck'
But I want to track it. So how do I do that using deep sort? There's no YOLO implementation in deep sort
Sorry @Flock1 i have no idea about it but I want to count vehicle(car,bus,motorbyce ) how can I count it using deep sort any I dea or suggestion Thanks
Hi, Is there any YOLO implementation in deep sort? I have trained a model consisting (a small video) on my custom dataset. I want to track few Objects will Deep Sort be useful for that?
@buzdarbalooch Follow this https://github.com/Qidian213/deep_sort_yolov3
Deep SORT uses an appearance descriptor that has been trained on a person re-identification dataset. You probably want to change that with something that has been trained on cars/trucks. If you have a dataset, the necessary steps would roughly look like this:
HI @nwojke
I have a problem where I need to track exact looking objects (Color, appearance, you can say I tying to track same model bus ), So a reidentification model is no help at all here
So I was thinking maybe I should put more weight on kalam filters, , I see that we can change that using
D=Lambda∗Dk+(1−Lambda)∗Da
But I was not able to find the exact location in code where I can change this, so that it takes more weight on Kalman Mahalanobis distance, rather than on cosine distance, please help with this
But is this only way, is there any other way i can solve tracking of a similar-looking object
is there a different way i should train my reidentification model ?
if anyone has any ideas or solved similar problems, Please please share Thanks
@nwojke hi i am trying to train the cosine metric learning model available in this repository https://github.com/nwojke/cosine_metric_learning with data set VeRi available in https://github.com/JDAI-CV/VeRidataset. I have followed some recommendations that you described in this repository https://github.com/nwojke/cosine_metric_learning/issues/5 carry out the training with this data set. I have tried to carry out the training with the data set as supplied without any modification in the files, however I have the following errors. `2020-07-05 21:43:25.232079: W tensorflow/core/framework/op_kernel.cc:1502] OP_REQUIRES failed at tensor_array_ops.cc:447 : Invalid argument: TensorArray map/TensorArray_1_0: Could not write to TensorArray index 1 because the value shape is [247,411,3] which is incompatible with the TensorArray's inferred element shape: [152,151,3] (consider setting infer_shape=False). EnqueueError: TensorArray map/TensorArray_1_0: Could not write to TensorArray index 1 because the value shape is [247,411,3] which is incompatible with the TensorArray's inferred element shape: [152,151,3] (consider setting infer_shape=False). [[node map/while/TensorArrayWrite/TensorArrayWriteV3 (defined at /content/gdrive/My Drive/Trainig_Cosenet/cosine_metric_learning-master/train_app.py:251) ]]
Errors may have originated from an input operation. Input Source operations connected to node map/while/TensorArrayWrite/TensorArrayWriteV3: map/while/DecodeJpeg (defined at /content/gdrive/My Drive/Trainig_Cosenet/cosine_metric_learning-master/train_app.py:250)
Original stack trace for 'map/while/TensorArrayWrite/TensorArrayWriteV3':
File "train_veri.py", line 133, in
I wanted to ask you how the file folders should be or what modifications I have to make in order to carry out this training. In addition, the training will be done from scratch, there is no pre-training model that can be used for the transfer learning. or in which part can I enter the weights obtained with the data set "Market-1501-v15.09.15" so that the training does not take long time.
tanks you
@nwojke Hi, I have some question on this issue, I want to ask that, is it possible to train the cosine metric model for both human and vehicle classes so that the deep sort model can track both human and car/trucks and etc? Have you ever tried to train different classes with combining multiple categories?
Hey guys, I want to know how can I use this to track trucks in a video. Should I replace this CNN with a CNN trained to detect trucks or re-train the CNN for trucks? Let me know