Hi! @kbrodt Thanks for sharing this great code！

Where are the codes of the evaluation metrics (for example: ADE, FDE, minADE, minFDE and so on)? Or where can I find it?

Looking forward to your reply！

Thanks for sharing this awesome codes!

Is it necessary to train the model for 120 epochs? Since there are more than 1M training samples. Can you share some performance during the training progress? Such as the performance with 30epochs, 60 epochs, 90 epochs? Since I trained it for several epochs but the loss is still very large.

To double check the training process, can you share how many training samples for each epoch?

Thanks so much!

Hi! I tried your method and I observed that in training, l2loss and log_softmax have so large difference. so my network does not learn multimodal tracks, only one best track is fitted. Do you have any solution?

Thank you for amazing works.

I found the conversion error at prerender.py

https://github.com/kbrodt/waymo-motion-prediction-2021/blob/3665d4b3ba39a6b879b663747f93b6e525018c00/prerender.py#L706 https://github.com/kbrodt/waymo-motion-prediction-2021/blob/3665d4b3ba39a6b879b663747f93b6e525018c00/prerender.py#L707

These two lines has conversion error.

In order to convert coordinate properly, it should be changed like this tmp[3] = other_v_yaw - ANGLE tmp[4] = other_bbox_yaw - ANGLE

Hi, @kbrodt,

Thank you for your great work. I am new to this area, so I have a question regarding to the lane type of waymo dataset. In general, the lanes in waymo dataset, can be broadly categorized into 6: lanecenters, roadlines, stopsign, speedbump, roadedge and crosswalks. So, is the lane centers are just the the center of each line in which the vehicles can drive on? Theoretically, a car's center should be on the lanecenters if this car is staying in this lane. Is my interpretation right? Also, is the lanecenters in waymo the same as the centerlines in Argoverse Dataset? Are you familiar with that dataset? I am looking forward to your reply. Thank you in advance.

Hello!

Thanks for sharing you work!

I have one questions: from the motion dataset, is that possible to know if two vehicles are driving on the same lane? I notice that there is a column called "roadgraph_samples/id" but I am not sure whether that means the lane ID.

Thanks.

Привет Кирилл,

Я вижу что ты разбираешься в машинном обучении. Я хотел бы сконнектиться и перенять опыт, если возможно. Как с тобой можно связаться?

Дмитрий

The infomation is showed as below: Additional permissions required to list objects in this bucket. Ask a bucket owner to grant you 'storage.objects.list' permission.

After running the following command :

(venvpy37cu10) [015214329@condo4 project]$ python train.py --train-data ./train --dev-data ./dev --save ./xception71 --model xception71 --img-res 224 --in-channels 25 --time-limit 80 --n-traj 6 --lr 0.001 --batch-size 48 --n-epochs 120

Below error has occurred :

Hi,

Thanks for your open-source work.

I don't understand the magic_const and shift in rasterize() function prerender.py.

Would you please give some explanation?

I tried running with all requirements , it keeps on reading the records but when it tries to write down .....here it fails , please help me to proceed further

root@6f4483fc53c1:/app/waymo-adas-main/waymo-motion-prediction-2021# python3 prerender.py --data /app/waymo-adas-main/waymo-dataset/original/validation/ --out /app/waymo-adas-main/data/train1
False
Namespace(data='/app/waymo-adas-main/waymo-dataset/original/validation/', each=0, n_jobs=20, n_shards=8, no_valid=False, out='/app/waymo-adas-main/data/train1', use_vectorize=False)
1215it [00:28, 42.50it/s]
  0%|                                                                                                                                                                                   | 0/1215 [00:00<?, ?it/s]
multiprocessing.pool.RemoteTraceback:
"""
Traceback (most recent call last):
  File "/usr/lib/python3.8/multiprocessing/pool.py", line 125, in worker
    result = (True, func(*args, **kwds))
  File "prerender.py", line 731, in merge
    parsed = tf.io.parse_single_example(data, features_description)
  File "/usr/local/lib/python3.8/dist-packages/tensorflow/python/util/traceback_utils.py", line 153, in error_handler
    raise e.with_traceback(filtered_tb) from None
  File "/usr/local/lib/python3.8/dist-packages/tensorflow/python/eager/execute.py", line 58, in quick_execute
    tensors = pywrap_tfe.TFE_Py_Execute(ctx._handle, device_name, op_name,
UnicodeDecodeError: 'utf-8' codec can't decode byte 0x96 in position 40: invalid start byte
"""

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "prerender.py", line 836, in <module>
    main()
  File "prerender.py", line 832, in main
    r.get()
  File "/usr/lib/python3.8/multiprocessing/pool.py", line 771, in get
    raise self._value
UnicodeDecodeError: 'utf-8' codec can't decode byte 0x96 in position 40: invalid start byte

Congrats! Actually I have a question which has confused me a few days. When we rasterize the map information, why we need to shift and rotate the local coordinate system that the target agent is located at a specific location? I konw that many papers has used this method like buliding a relative coordinate system and the center is target agent. why? and what is the meaning of "to eliminate the redundant degrees of freedom" in your report? Thank you in advance!