Faster-rcnn.pytorch: TypeError: can't assign a numpy.int64 to a torch.FloatTensor

Created on 28 Feb 2019 · 15Comments · Source: jwyang/faster-rcnn.pytorch

Guys,
Did anyone know what caused this issue below? Many thanks in advance.
Torch: 0.40
Python: 3.6

We use voc format of KITTI dataset. The only thing I changed for the original code from this repo is to change the picture format from "jpg" to "png" which is the format for KITTI dataset. When I run ''python trainvel_net.py", I got error below:

Traceback (most recent call last):
File "trainval_net.py", line 209, in
imdb.num_classes, training=True)
File "/home/NewPartion/pycharm/faster-rcnn.pytorch/lib/roi_data_layer/roibatchLoader.py", line 54, in __init__
self.ratio_list_batch[left_idx:(right_idx+1)] = target_ratio
TypeError: can't assign a numpy.int64 to a torch.FloatTensor

Source

goodproj13

Most helpful comment

self.ratio_list_batch[left_idx:(right_idx+1)] = torch.tensor(target_ratio.astype(np.float64)) # trainset ratio list ,each batch is same number
This fixed the issue for 0.4.0

benedictflorance on 6 Jun 2019

🎉4 ❤1 👍1

All 15 comments

please paste all the output from your training run.. so we can see if you enabled CUDA, etc...

EMCP on 14 Mar 2019

Did you solve this problem? I have the same issue ...

13952522076 on 12 Apr 2019

I didn't solve the problem and I ended up using other's. Sorry!

From: Xu Ma notifications@github.com
Sent: Friday, April 12, 2019 11:26:30 AM
To: jwyang/faster-rcnn.pytorch
Cc: Dongfang Liu; Author
Subject: Re: [jwyang/faster-rcnn.pytorch] TypeError: can't assign a numpy.int64 to a torch.FloatTensor (#452)

Did you solve this problem? I have the same issue ...

—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHubhttps://github.com/jwyang/faster-rcnn.pytorch/issues/452#issuecomment-482616857, or mute the threadhttps://github.com/notifications/unsubscribe-auth/ALO7KPSBVIN5LS4PTJWHQJDPQCS2NANCNFSM4G25NUQQ.

goodproj13 on 12 Apr 2019

Maybe I should try mmdetection or Detectron ...

13952522076 on 12 Apr 2019

Change original code to
temp = torch.ones(batch_size)*target_ratio
self.ratio_list_batch[left_idx:(right_idx+1)] = temp

Weizhongjin on 16 Apr 2019

@Weizhongjin

Change original code to
temp = torch.ones(batch_size)*target_ratio
self.ratio_list_batch[left_idx:(right_idx+1)] = temp

Doing the above fix throws this error:

**TypeError: mul() received an invalid combination of arguments - got (numpy.int64), but expected one of:
 * (Tensor other)
      didn't match because some of the arguments have invalid types: (numpy.int64)
 * (float other)
      didn't match because some of the arguments have invalid types: (numpy.int64)

**

Has anyone found a fix for this?

benedictflorance on 4 Jun 2019

@EMCP

Called with args:
Namespace(batch_size=1, checkepoch=1, checkpoint=0, checkpoint_interval=10000, checksession=1, class_agnostic=False, cuda=True, dataset='hollywoodheads_scuta', disp_interval=100, lamda=0.1, large_scale=False, lr=0.002, lr_decay_gamma=0.1, lr_decay_step=6, mGPUs=True, max_epochs=5, net='vgg16', num_workers=0, optimizer='sgd', resume=False, save_dir='data/adaptation/experiments', session=1, start_epoch=1, use_tfboard=False)
loading our dataset...........
/export/livia/home/vision/bflorance/da-faster-rcnn-PyTorch/lib/model/utils/config.py:376: YAMLLoadWarning: calling yaml.load() without Loader=... is deprecated, as the default Loader is unsafe. Please read https://msg.pyyaml.org/load for full details.
  yaml_cfg = edict(yaml.load(f))
Using config:
{'ANCHOR_RATIOS': [0.5, 1, 2],
 'ANCHOR_SCALES': [4, 8, 16, 32],
 'CROP_RESIZE_WITH_MAX_POOL': False,
 'CUDA': False,
 'DATA_DIR': '/export/livia/home/vision/bflorance/da-faster-rcnn-PyTorch/data',
 'DEDUP_BOXES': 0.0625,
 'DSN_DIFF_WEIGHT': 100000,
 'EPS': 1e-14,
 'EXP_DIR': 'vgg16',
 'FEAT_STRIDE': [16],
 'GPU_ID': 0,
 'MATLAB': 'matlab',
 'MAX_NUM_GT_BOXES': 50,
 'MOBILENET': {'DEPTH_MULTIPLIER': 1.0,
               'FIXED_LAYERS': 5,
               'REGU_DEPTH': False,
               'WEIGHT_DECAY': 4e-05},
 'PIXEL_MEANS': array([[[102.9801, 115.9465, 122.7717]]]),
 'POOLING_MODE': 'align',
 'POOLING_SIZE': 7,
 'RESNET': {'FIXED_BLOCKS': 1, 'MAX_POOL': False},
 'RNG_SEED': 3,
 'ROOT_DIR': '/export/livia/home/vision/bflorance/da-faster-rcnn-PyTorch',
 'TEST': {'BBOX_REG': True,
          'HAS_RPN': True,
          'MAX_SIZE': 1000,
          'MODE': 'nms',
          'NMS': 0.3,
          'PROPOSAL_METHOD': 'gt',
          'RPN_MIN_SIZE': 16,
          'RPN_NMS_THRESH': 0.7,
          'RPN_POST_NMS_TOP_N': 300,
          'RPN_PRE_NMS_TOP_N': 6000,
          'RPN_TOP_N': 5000,
          'SCALES': [600],
          'SVM': False},
 'TRAIN': {'ASPECT_GROUPING': False,
           'BATCH_SIZE': 256,
           'BBOX_INSIDE_WEIGHTS': [1.0, 1.0, 1.0, 1.0],
           'BBOX_NORMALIZE_MEANS': [0.0, 0.0, 0.0, 0.0],
           'BBOX_NORMALIZE_STDS': [0.1, 0.1, 0.2, 0.2],
           'BBOX_NORMALIZE_TARGETS': True,
           'BBOX_NORMALIZE_TARGETS_PRECOMPUTED': True,
           'BBOX_REG': True,
           'BBOX_THRESH': 0.5,
           'BG_THRESH_HI': 0.5,
           'BG_THRESH_LO': 0.0,
           'BIAS_DECAY': False,
           'BN_TRAIN': False,
           'DISPLAY': 10,
           'DOUBLE_BIAS': True,
           'FG_FRACTION': 0.25,
           'FG_THRESH': 0.5,
           'GAMMA': 0.1,
           'HAS_RPN': True,
           'IMS_PER_BATCH': 1,
           'LEARNING_RATE': 0.01,
           'MAX_SIZE': 1000,
           'MOMENTUM': 0.9,
           'PROPOSAL_METHOD': 'gt',
           'RPN_BATCHSIZE': 256,
           'RPN_BBOX_INSIDE_WEIGHTS': [1.0, 1.0, 1.0, 1.0],
           'RPN_CLOBBER_POSITIVES': False,
           'RPN_FG_FRACTION': 0.5,
           'RPN_MIN_SIZE': 8,
           'RPN_NEGATIVE_OVERLAP': 0.3,
           'RPN_NMS_THRESH': 0.7,
           'RPN_POSITIVE_OVERLAP': 0.7,
           'RPN_POSITIVE_WEIGHT': -1.0,
           'RPN_POST_NMS_TOP_N': 2000,
           'RPN_PRE_NMS_TOP_N': 12000,
           'SCALES': [600],
           'SNAPSHOT_ITERS': 5000,
           'SNAPSHOT_KEPT': 3,
           'SNAPSHOT_PREFIX': 'res101_faster_rcnn',
           'STEPSIZE': [30000],
           'SUMMARY_INTERVAL': 180,
           'TRIM_HEIGHT': 600,
           'TRIM_WIDTH': 600,
           'TRUNCATED': False,
           'USE_ALL_GT': True,
           'USE_FLIPPED': True,
           'USE_GT': False,
           'WEIGHT_DECAY': 0.0005},
 'USE_GPU_NMS': True}
Loaded dataset `hollywoodheads_scuta_2007_train_s` for training
Set proposal method: gt
Appending horizontally-flipped training examples...
hollywoodheads_scuta_2007_train_s gt roidb loaded from /export/livia/home/vision/bflorance/da-faster-rcnn-PyTorch/data/cache/hollywoodheads_scuta_2007_train_s_gt_roidb.pkl
done
Preparing training data...
done
before filtering, there are 200 images...
after filtering, there are 200 images...
Source Train Size =  200
Loaded dataset `hollywoodheads_scuta_2007_train_t` for training
Set proposal method: gt
Appending horizontally-flipped training examples...
hollywoodheads_scuta_2007_train_t gt roidb loaded from /export/livia/home/vision/bflorance/da-faster-rcnn-PyTorch/data/cache/hollywoodheads_scuta_2007_train_t_gt_roidb.pkl
done
Preparing training data...
done
before filtering, there are 200 images...
after filtering, there are 200 images...
Target Train Size =  200
source 200 target 200 roidb entries
Traceback (most recent call last):
  File "da_trainval_net.py", line 257, in <module>
    s_imdb.num_classes, training=True)
  File "/export/livia/home/vision/bflorance/da-faster-rcnn-PyTorch/lib/roi_da_data_layer/roibatchLoader.py", line 55, in __init__
    self.ratio_list_batch[left_idx:(right_idx+1)] = target_ratio    # trainset ratio list ,each batch is same number
TypeError: can't assign a numpy.int64 to a torch.FloatTensor

benedictflorance on 4 Jun 2019

Can you possibly upgrade to PyTorch 1.0 ? I've used exclusively 1.x and had zero issues for months now

EMCP on 4 Jun 2019

@EMCP I'm working on domain adaptation (https://github.com/tiancity-NJU/da-faster-rcnn-PyTorch) and it uses the PyTorch 0.4 version of this repo. :(

benedictflorance on 4 Jun 2019

okay i am abroad from my deep learning rig, so I cannot test PyTorch .4 until mid June... but I would diff the two versions and ensure you've got any/all bug fixes pushed to the 1.x version... I never ran the Master branch so I cannot get a sense if it's been properly patched or not..

IMO , the repo owner should just do a release branch for older PyTorch versions and stick to the bleeding edge in Master root, instead of this 1.x off on the side strategy.

EMCP on 4 Jun 2019

self.ratio_list_batch[left_idx:(right_idx+1)] = torch.tensor(target_ratio.astype(np.float64)) # trainset ratio list ,each batch is same number
This fixed the issue for 0.4.0

benedictflorance on 6 Jun 2019

🎉4 ❤1 👍1

feel free to submit a PR and close @benedictflorance

EMCP on 8 Jun 2019

Yeah, I've submitted one.
https://github.com/jwyang/faster-rcnn.pytorch/pull/573

benedictflorance on 9 Jun 2019

👍1

/lib/roi_data_layer/roibatchLoader.py
line 52, target_ratio = 1 change to:
target_ratio = np.array(1)

ljk1072911239 on 20 Jun 2019

Solving this definitely, in this line, change:

ratio_large = 2 # largest ratio to preserve.

to:

ratio_large = 2.0 # largest ratio to preserve.

PyTorch is creating a tensor with these values and its type is inferred from the data. A value of 2 will be inferred as int, thus, changing to a floating point value will fix it.
Note that this error only occurs when it falls in this ifstatement. The below if statement uses ratio = ratio_small where ratio_small = 0.5 is already a floating point value as defined in the beginning of the function.