megvii-model/crowddetection

crowddetection's Issues

Erro when compiling

Hi, when i make sh ./setup.sh i get this error

nms.cu:1:10: fatal error: megbrain_pubapi.h: No such file or directory
#include "megbrain_pubapi.h"
^~~~~~~~~~~~~~~~~~~
compilation terminated.

How can i fix it?

set_cpu_nms

To understand the set NMS operation, I run the sub main func in utils/set_nms_utils , changing keep = py_cpu_nms(boxes,nms_thresh) as keep = set_cpu_nms(boxes,nms_thresh) , the ERROR encounters as follow:
"set_nms_utils.py", line 29, in set_cpu_nms
numbers = dets[:, 5]
IndexError: index 5 is out of bounds for axis 1 with size 5
Please Help solve this \ ( * ^ * ) /

Errors when runing on myself dataset

Environment

• CUDA 10.2
• megengine 3.1
• python 3.6

Hi, I run this code based on the crowd person dataset, It seem everything is ok.

11 23:56:11 e0, 112/3750, lr:0.002123, total_loss:0.870193, rpn_cls:0.212023, rpn_loc:0.182936, rcnn_emd:0.475234
11 23:56:18 e0, 113/3750, lr:0.002127, total_loss:0.817391, rpn_cls:0.129117, rpn_loc:0.066281, rcnn_emd:0.621992
11 23:56:19 e0, 114/3750, lr:0.002132, total_loss:1.136824, rpn_cls:0.205079, rpn_loc:0.195120, rcnn_emd:0.736625

But, when I try to run this code based on myself dataset, it always occur the error as follows:

 File "/root/ld/PycharmProjects/CrowdDetection/model/emd_simple/train.py", line 60, in train_one_epoch
    losses = propagate()
megengine._internal.exc.MegBrainError: MegBrain core throws exception: mgb::MegDNNError
bad input shape for polyadic operator: {2034,12}, {2034,4}
| Associated operator: id=130407 name=SUB(reshape[130071],reshape[130401])[130407] type=mgb::opr::Elemwise
|   input variables: 
|     0: {id:130072, shape:{2034,12}, Float32, owner:reshape(concat[130063])[130071]{Reshape}, name:reshape(concat[130063])[130071], slot:0, gpu0:0, d, 8, 8}
|     1: {id:130402, shape:{2034,4}, Float32, owner:reshape(indexing_multi_axis_vec[130398])[130401]{Reshape}, name:reshape(indexing_multi_axis_vec[130398])[130401], slot:0, gpu0:0, d, 8, 8}
|   output variables: 
|     0: {id:130408, shape:{}, Float32, owner:SUB(reshape[130071],reshape[130401])[130407]{Elemwise}, name:SUB(reshape[130071],reshape[130401])[130407], slot:0, gpu0:0, d, 8, 8}

Do you have any suggestion about that?
Thanks

Failed to request for multiple GPUs during inference

test.py provides API for multi-GPUs testing. However, when I set -d 4, the program seems to request memory on only GPU 0 which leads to OOM.

11 16:42:50[mgb] ERR cudaMalloc failed while requesting 57933824 bytes (55.250MiB) of memory; error: out of memory(last_err=2(out of memory) device=0 mem_free=29.312MiB mem_tot=24220.312MiB)
11 16:42:50[mgb] could not allocate memory on device 0; try to gather free blocks from child streams, got 0.00MiB(0 bytes).

Have you met this problem before?

Problem with @jit.trace(symbolic=True) in the train.py of the cascade_emd model

When I run the cascade_emd model, I met the error as the following. I appreciate it if you could help me out. Thank you in advance.

Traceback (most recent call last):
File "train.py", line 167, in
run_train()
File "train.py", line 164, in run_train
train(args)
File "train.py", line 156, in train
worker(0, 1, args)
File "train.py", line 119, in worker
train_one_epoch(model, train_loader, opt, max_steps, rank, epoch_id, gpu_num)
File "train.py", line 58, in train_one_epoch
losses = propagate()
File "/home/xinmiao/anaconda3/envs/CRDET/lib/python3.6/site-packages/megengine/jit/init.py", line 424, in call
self._compiled_func()
File "/home/xinmiao/anaconda3/envs/CRDET/lib/python3.6/site-packages/megengine/_internal/mgb.py", line 1208, in call
self._execute()
File "/home/xinmiao/anaconda3/envs/CRDET/lib/python3.6/site-packages/megengine/_internal/mgb.py", line 1092, in _execute
return _mgb.AsyncExec__execute(self)
megengine._internal.exc.MegBrainError: MegBrain core throws exception: mgb::AssertionError
assertion `begin >= 0 && end >= begin && end <= size_ax' failed at /home/code/src/core/impl/tensor.cpp:151: mgb::SubTensorSpec mgb::Slice::apply(megdnn::TensorLayout, int) const
extra message: index out of bound: layout={511(1),1(1)}; request begin=None end=2 step=None axis=1

• bt:/home/xinmiao/anaconda3/envs/CRDET/lib/python3.6/site-packages/megengine/_internal/_mgb.cpython-36m-x86_64-linux-gnu.so{1e36052,1edec06,1fc6782,1fc6fd0}
  | Associated operator: id=160315 name=subtensor(argsort[160305]:o0)[160315] type=mgb::opr::Subtensor
  | input variables:
  | 0: {id:160306, shape:{511,1}, Float32, owner:argsort(MUL[160303])[160305]{ArgsortForward}, name:argsort(MUL[160303])[160305]:o0, slot:0, gpu0:0, d, 8, 1}
  | 1: {id:21, shape:{1}, Int32, owner:2[20]{ImmutableTensor}, name:2[20], slot:0, gpu0:0, s, 2, 2}
  | output variables:
  | 0: {id:160316, shape:{553,2}, Float32, owner:subtensor(argsort[160305]:o0)[160315]{Subtensor}, name:subtensor(argsort[160305]:o0)[160315], slot:0, gpu0:0, d, 8, 8}
  |
  | Unoptimized equivalent of associated operator: id=10623 name=subtensor(argsort[10615]:o0)[10623] type=mgb::opr::Subtensor
  | input variables:
  | 0: {id:10616, shape:{}, Float32, owner:argsort(MUL[10611])[10615]{ArgsortForward}, name:argsort(MUL[10611])[10615]:o0, slot:0, gpu0:0, d, 8, 1}
  | 1: {id:21, shape:{1}, Int32, owner:2[20]{ImmutableTensor}, name:2[20], slot:0, gpu0:0, s, 2, 2}
  | output variables:
  | 0: {id:10624, shape:{}, Float32, owner:subtensor(argsort[10615]:o0)[10623]{Subtensor}, name:subtensor(argsort[10615]:o0)[10623], slot:0, gpu0:0, d, 8, 8}

请问代码支持断点训练吗？

我想从某个epoch继续训练，但是输入-r参数，它还是从头开始训练的

Difference between EMD loss and smoothL1?

Your implementation of EMD loss here seems to be the same as smoothL1 loss between anchor and its pred boxes.What's the difference them?

So is it AP50 rather than coco-style AP in the paper?