search

skyhehe123 / SA-SSD

SA-SSD: Structure Aware Single-stage 3D Object Detection from Point Cloud (CVPR 2020)
492 stars 106 forks source link

error in multi-class training:'Tensor' object has no attribute 'bool' #61

Open rolan123 opened 4 years ago

rolan123 commented 4 years ago

error in multi-class training: Traceback (most recent call last): File "train.py", line 127, in main() File "train.py", line 117, in main log_interval = cfg.log_config.interval File "/root/aaaaaaaa/SA-SSD/tools/train_utils/init.py", line 99, in train_model log_interval = log_interval File "/root/aaaaaaaa/SA-SSD/tools/train_utils/init.py", line 57, in train_one_epoch outputs = batch_processor(model, data_batch) File "/root/aaaaaaaa/SA-SSD/tools/train_utils/init.py", line 29, in batch_processor losses = model(data) File "/root/miniconda3/envs/SASSD/lib/python3.6/site-packages/torch/nn/modules/module.py", line 493, in call result = self.forward(*input, *kwargs) File "/root/miniconda3/envs/SASSD/lib/python3.6/site-packages/torch/nn/parallel/data_parallel.py", line 152, in forward outputs = self.parallel_apply(replicas, inputs, kwargs) File "/root/miniconda3/envs/SASSD/lib/python3.6/site-packages/torch/nn/parallel/data_parallel.py", line 162, in parallel_apply return parallel_apply(replicas, inputs, kwargs, self.device_ids[:len(replicas)]) File "/root/miniconda3/envs/SASSD/lib/python3.6/site-packages/torch/nn/parallel/parallel_apply.py", line 83, in parallel_apply raise output File "/root/miniconda3/envs/SASSD/lib/python3.6/site-packages/torch/nn/parallel/parallel_apply.py", line 59, in _worker output = module(input, kwargs) File "/root/miniconda3/envs/SASSD/lib/python3.6/site-packages/torch/nn/modules/module.py", line 493, in call result = self.forward(*input, kwargs) File "/root/aaaaaaaa/SA-SSD/mmdet/models/detectors/base.py", line 79, in forward return self.forward_train(img, img_meta, kwargs) File "/root/aaaaaaaa/SA-SSD/mmdet/models/detectors/single_stage.py", line 97, in forward_train ret['anchors_mask'], ret['gt_bboxes'], ret['gt_labels'], thr=self.train_cfg.rpn.anchor_thr) File "/root/aaaaaaaa/SA-SSD/mmdet/models/single_stage_heads/ssd_rotate_head.py", line 361, in get_guided_anchors opp_labels = (box_preds[..., -1] > 0) ^ dir_labels.bool() AttributeError: 'Tensor' object has no attribute 'bool'

How to deal with it?

rolan123 commented 4 years ago

my pytorch version=1.1.0

PengfeiMa commented 4 years ago

try opp_labels = (box_preds[..., -1] > 0) ^ dir_labels.bool() ----> opp_labels = (box_preds[..., -1] > 0) ^ dir_labels.byte()

rolan123 commented 4 years ago

@PengfeiMa Thank you, the error in multi-class training has been resolved. I did the training and the evaluation, and I found a new problem： Because the few number of cyclists and the training had been over-fit, the AP began to decline. But at the same time car's AP can continue to improve. Cars are about twenty times as numerous as cyclists.

How to deal with it?

PengfeiMa commented 4 years ago

@rolan123 What's your final results on per class?

rolan123 commented 4 years ago

epoch=70 Car AP@0.70, 0.70, 0.70: bbox AP:97.99, 90.07, 89.72 bev AP:98.49, 89.36, 88.96 3d AP:90.10, 86.40, 86.47 aos AP:97.94, 89.88, 89.28 Car AP@0.70, 0.50, 0.50: bbox AP:97.99, 90.07, 89.72 bev AP:98.07, 90.06, 89.78 3d AP:98.04, 90.06, 89.77 aos AP:97.94, 89.88, 89.28 Pedestrian AP@0.50, 0.50, 0.50: bbox AP:64.24, 56.45, 55.70 bev AP:63.57, 55.13, 54.11 3d AP:61.72, 53.77, 46.88 aos AP:60.37, 52.79, 51.84 Pedestrian AP@0.50, 0.25, 0.25: bbox AP:64.24, 56.45, 55.70 bev AP:75.22, 66.63, 58.93 3d AP:74.84, 66.39, 58.78 aos AP:60.37, 52.79, 51.84 Cyclist AP@0.50, 0.50, 0.50: bbox AP:89.66, 84.00, 82.35 bev AP:87.88, 80.72, 78.53 3d AP:87.29, 79.94, 77.71 aos AP:89.59, 83.48, 81.79 Cyclist AP@0.50, 0.25, 0.25: bbox AP:89.66, 84.00, 82.35 bev AP:89.05, 85.74, 79.35 3d AP:89.04, 85.74, 79.35 aos AP:89.59, 83.48, 81.79

epoch=80: Car AP@0.70, 0.70, 0.70: bbox AP:98.24, 90.19, 89.81 bev AP:97.79, 89.50, 89.08 3d AP:90.12, 86.85, 86.61 aos AP:98.21, 90.02, 89.55 Car AP@0.70, 0.50, 0.50: bbox AP:98.24, 90.19, 89.81 bev AP:98.28, 90.18, 89.85 3d AP:98.25, 90.17, 89.83 aos AP:98.21, 90.02, 89.55 Pedestrian AP@0.50, 0.50, 0.50: bbox AP:63.58, 62.11, 55.34 bev AP:63.02, 54.94, 53.83 3d AP:61.89, 53.77, 52.48 aos AP:61.06, 58.98, 52.45 Pedestrian AP@0.50, 0.25, 0.25: bbox AP:63.58, 62.11, 55.34 bev AP:74.64, 66.71, 65.34 3d AP:74.73, 66.70, 65.33 aos AP:61.06, 58.98, 52.45 Cyclist AP@0.50, 0.50, 0.50: bbox AP:89.08, 88.90, 82.40 bev AP:86.78, 85.01, 78.72 3d AP:81.74, 80.38, 78.26 aos AP:88.99, 88.63, 82.14 Cyclist AP@0.50, 0.25, 0.25: bbox AP:89.08, 88.90, 82.40 bev AP:87.58, 85.69, 79.43 3d AP:87.58, 85.69, 79.43 aos AP:88.99, 88.63, 82.14

PengfeiMa commented 4 years ago

@rolan123 I test my epoch-70 and epoch-80, the results are almost the same as yours.