Add a column of data

[xuV2]

Hello! @pangsu0613 I read your paper(Fast-CLOCs), and I also added a column of data to your code, changing it from 1p4 to 1p5, but there is an error in my code( self.opt.step() in fastai_optim.py).RuntimeError: The size of tensor a (4) must match the size of tensor b (5) at non-singleton dimension 1.

[pangsu0613]

Hello @xuV2 . If one needs to add more channels (or modify the number of channels), these places below need to be modified:
(1)you also need to modify the build_stage2_training function below, add another colum in variable "overlaps".

CLOCs/second/utils/eval.py

Line 127 in cad14fd

def build_stage2_training(boxes, query_boxes, criterion, scores_3d, scores_2d, dis_to_lidar_3d,overlaps,tensor_index):

(2)change the hardcoded "4" channels into "5", from line 521-527 (or other number of channels if you want):

CLOCs/second/pytorch/models/voxelnet.py

Line 503 in cad14fd

overlaps1 = np.zeros((900000,4),dtype=box_2d_preds.detach().cpu().numpy().dtype)

(3)change the hardcoded "4" into "5" in CLOCs model:

CLOCs/second/pytorch/models/fusion.py

Line 27 in cad14fd

nn.Conv2d(4,18,1),

My apology to hardcoded these information. I'll update the repo later to remove these hardcoded value.

[xuV2]

Hello! @pangsu0613 First of all, thank you for your prompt reply. But I have changed the code you mentioned, but when the code runs to train.py(line 279) during training, it will still make mistakes.When I delete 279 lines, the code can be trained.
CLOCs/blob/master/second/pytorch/train.py
279 mixed_optimizer.step()

[xuV2]

The reason for the error is
RuntimeError: The size of tensor a (4) must match the size of tensor b (5) at non-singleton dimension 1

[pangsu0613]

mixed_optimizer.step() should not be deleted. Could you show me the whole error message?

[xuV2]

Traceback (most recent call last):
File "./pytorch/train.py", line 985, in
fire.Fire()
File "/root/anaconda3/envs/CLOCs/lib/python3.6/site-packages/fire/core.py", line 141, in Fire
component_trace = _Fire(component, args, parsed_flag_args, context, name)
File "/root/anaconda3/envs/CLOCs/lib/python3.6/site-packages/fire/core.py", line 471, in Fire
target=component.name)
File "/root/anaconda3/envs/CLOCs/lib/python3.6/site-packages/fire/core.py", line 681, in CallAndUpdateTrace
component = fn(*varargs, **kwargs)
File "./pytorch/train.py", line 431, in train
raise e
File "./pytorch/train.py", line 336, in train
mixed_optimizer.step() # 这句话有问题 # 通过梯度下降更新参数
File "/root/CLOCs/torchplus/train/fastai_optim.py", line 136, in step
self.opt.step() # 有问题
File "/root/anaconda3/envs/CLOCs/lib/python3.6/site-packages/torch/optim/adam.py", line 93, in step
exp_avg.mul(beta1).add(1 - beta1, grad)
RuntimeError: The size of tensor a (4) must match the size of tensor b (5) at non-singleton dimension 1

[pangsu0613]

did you change the "4" at line 521 and line 524 in voxelnet.py?

[xuV2]

iou_test_tensor = iou_test_tensor.reshape(1, 5, 1, 900000) # 原来是4
# print(iou_test_tensor.shape) (1, 5, 1, 900000)
tensor_index_tensor = tensor_index_tensor.reshape(-1, 2)
if max_num == 0:
non_empty_iou_test_tensor = torch.zeros(1, 5, 1, 2)
non_empty_iou_test_tensor[:, :, :, :] = -1
non_empty_tensor_index_tensor = torch.zeros(2, 2)
non_empty_tensor_index_tensor[:, :] = -1
else:
non_empty_iou_test_tensor = iou_test_tensor[:, :, :, :max_num]
# print(non_empty_iou_test_tensor.shape) # torch.Size([1, 5, 1, 313987])
non_empty_tensor_index_tensor = tensor_index_tensor[:max_num, :]

[pangsu0613]

could you delete all the files in the working directory (including the adam_optimizer-xxxx.tckpt and other files)?

[xuV2]

I'm sorry I don't have this ability. I'm just a novice.😭

[pangsu0613]

when you run training in the terminal:
python ./pytorch/train.py train --config_path=./configs/car.fhd.config --model_dir=/dir/to/your_model_dir
you create a model directory (--model_dir=/dir/to/your_model_dir), just delete everything in that directory.

[xuV2]

感谢您的建议，我已经成功解决了我的问题，真的非常感谢您的建议！