KeyError: 'boxes'

[zwyyy215]

(OICR) mickey@a208-System-Product-Name:~/OICR-pytorch$ CUDA_VISIBLE_DEVICES=0 python3 trainval_net.py Called with args: Namespace(batch_size=2, checkpoint=-1, checkpoint_interval=500, class_agnostic=False, cuda=False, dataset='pascal_voc', disp_interval=50, large_scale=False, load_dir=None, lr=0.001, lr_decay_gamma=0.1, lr_decay_step=5, mGPUs=False, max_epochs=20, model='oicr', net='vgg16', num_workers=4, optimizer='sgd', save_dir='output', session=1, start_epoch=1, threshold=0.01, use_tb=False, vis=False) /home/mickey/OICR-pytorch/lib/model/utils/config.py:392: 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': [8, 16, 32], 'CROP_RESIZE_WITH_MAX_POOL': False, 'CUDA': False, 'DATA_DIR': '/home/mickey/OICR-pytorch/data', 'DEDUP_BOXES': 0.125, 'EPS': 1e-14, 'EXP_DIR': 'vgg16', 'FEAT_STRIDE': [16], 'GPU_ID': 0, 'MATLAB': 'matlab', 'MAX_NUM_GT_BOXES': 20, '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': 2, 'ROOT_DIR': '/home/mickey/OICR-pytorch', 'TEST': {'BBOX_REG': True, 'HAS_RPN': True, 'MAX_SIZE': 2000, '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': [480, 576, 688, 864, 1200], 'SVM': False, 'USE_FLIPPED': True}, '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': 2, 'LEARNING_RATE': 0.01, 'MAX_IMAGE_SIZE': 2000, 'MAX_ROIS_SIZE': 6000, 'MAX_SIZE': 2000, 'MOMENTUM': 0.9, 'PROPOSAL_LIMIT': 2000, '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': [480, 576, 688, 864, 1200], 'SNAPSHOT_ITERS': 5000, 'SNAPSHOT_KEPT': 3, 'SNAPSHOT_PREFIX': 'res101_faster_rcnn', 'STEPSIZE': [40000], 'SUMMARY_INTERVAL': 180, 'TRIM_HEIGHT': 600, 'TRIM_WIDTH': 600, 'TRUNCATED': False, 'USE_ALL_GT': True, 'USE_FLIPPED': True, 'USE_GT': False, 'USE_SEG': True, 'WEIGHT_DECAY': 0.0005}, 'USE_GPU_NMS': True} WARNING: You have a CUDA device, so you should probably run with --cuda gt Loaded dataset voc_2007_trainval for training Set proposal method: gt Appending horizontally-flipped training examples... wrote gt roidb to /home/mickey/OICR-pytorch/data/cache/voc_2007_trainval_gt_roidb.pkl Traceback (most recent call last): File "trainval_net.py", line 231, in imdb, roidb, ratio_list, ratio_index = combined_roidb(args.imdb_name) File "/home/mickey/OICR-pytorch/lib/roi_data_layer/roidb.py", line 100, in combined_roidb roidbs = [get_roidb(s) for s in imdb_names.split('+')] File "/home/mickey/OICR-pytorch/lib/roi_data_layer/roidb.py", line 100, in roidbs = [get_roidb(s) for s in imdb_names.split('+')] File "/home/mickey/OICR-pytorch/lib/roi_data_layer/roidb.py", line 97, in get_roidb roidb = get_training_roidb(imdb) File "/home/mickey/OICR-pytorch/lib/roi_data_layer/roidb.py", line 82, in get_training_roidb imdb.append_flipped_images() File "/home/mickey/OICR-pytorch/lib/datasets/imdb.py", line 117, in append_flipped_images boxes = self.roidb[i]['boxes'].copy() KeyError: 'boxes'

[zwyyy215]

I changed __C.TRAIN.PROPOSAL_METHOD = 'gt' In vgg16.yml ,I changed PROPOSAL_METHOD:gt

[zwyyy215]

I don't know how to solve KeyError:'boxes', thank you

[jd730]

Hi @zwyyy215, The gt flag is a legacy from faster-RCNN. The reason why you got a key error is, roidb of gt is composed of labels and seg (this).

For weakly supervised object detection, we should not use gt bounding boxes. However, if you really want to use gt boxes, you can simply change above code by using how annots.pkl is made, maybe changing parse_rec. (this).

Thank you.

[zwyyy215]

Thank you for your guidance. When I change to selective_search, (OICR) mickey@a208-System-Product-Name:~/OICR-pytorch$ CUDA_VISIBLE_DEVICES=0 python 3 trainval_net.py Called with args: Namespace(batch_size=2, checkpoint=-1, checkpoint_interval=500, class_agnostic=False , cuda=False, dataset='pascal_voc', disp_interval=50, large_scale=False, load_dir=No ne, lr=0.001, lr_decay_gamma=0.1, lr_decay_step=5, mGPUs=False, max_epochs=20, model ='oicr', net='vgg16', num_workers=4, optimizer='sgd', save_dir='output', session=1, start_epoch=1, threshold=0.01, use_tb=False, vis=False) /home/mickey/OICR-pytorch/lib/model/utils/config.py:392: YAMLLoadWarning: calling ya ml.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': [8, 16, 32], 'CROP_RESIZE_WITH_MAX_POOL': False, 'CUDA': False, 'DATA_DIR': '/home/mickey/OICR-pytorch/data', 'DEDUP_BOXES': 0.125, 'EPS': 1e-14, 'EXP_DIR': 'vgg16', 'FEAT_STRIDE': [16], 'GPU_ID': 0, 'MATLAB': 'matlab', 'MAX_NUM_GT_BOXES': 20, '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': 2, 'ROOT_DIR': '/home/mickey/OICR-pytorch', 'TEST': {'BBOX_REG': True, 'HAS_RPN': True, 'MAX_SIZE': 2000, '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': [480, 576, 688, 864, 1200], 'SVM': False, 'USE_FLIPPED': True}, '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': 2, 'LEARNING_RATE': 0.01, 'MAX_IMAGE_SIZE': 2000, 'MAX_ROIS_SIZE': 6000, 'MAX_SIZE': 2000, 'MOMENTUM': 0.9, 'PROPOSAL_LIMIT': 2000, 'PROPOSAL_METHOD': 'selective_search', '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': [480, 576, 688, 864, 1200], 'SNAPSHOT_ITERS': 5000, 'SNAPSHOT_KEPT': 3, 'SNAPSHOT_PREFIX': 'res101_faster_rcnn', 'STEPSIZE': [40000], 'SUMMARY_INTERVAL': 180, 'TRIM_HEIGHT': 600, 'TRIM_WIDTH': 600, 'TRUNCATED': False, 'USE_ALL_GT': True, 'USE_FLIPPED': True, 'USE_GT': False, 'USE_SEG': True, 'WEIGHT_DECAY': 0.0005}, 'USE_GPU_NMS': True} WARNING: You have a CUDA device, so you should probably run with --cuda selective_search Loaded dataset voc_2007_trainval for training Set proposal method: selective_search Appending horizontally-flipped training examples... voc_2007_trainval gt roidb loaded from /home/mickey/OICR-pytorch/data/cache/voc_2007 _trainval_gt_roidb.pkl Traceback (most recent call last): File "trainval_net.py", line 231, in imdb, roidb, ratio_list, ratio_index = combined_roidb(args.imdb_name) File "/home/mickey/OICR-pytorch/lib/roi_data_layer/roidb.py", line 100, in combine d_roidb roidbs = [get_roidb(s) for s in imdb_names.split('+')] File "/home/mickey/OICR-pytorch/lib/roi_data_layer/roidb.py", line 100, in roidbs = [get_roidb(s) for s in imdb_names.split('+')] File "/home/mickey/OICR-pytorch/lib/roi_data_layer/roidb.py", line 97, in get_roid b roidb = get_training_roidb(imdb) File "/home/mickey/OICR-pytorch/lib/roi_data_layer/roidb.py", line 82, in get_trai ning_roidb imdb.append_flipped_images() File "/home/mickey/OICR-pytorch/lib/datasets/imdb.py", line 117, in append_flipped _images boxes = self.roidb[i]['boxes'].copy() File "/home/mickey/OICR-pytorch/lib/datasets/imdb.py", line 70, in roidb self._roidb = self.roidb_handler() File "/home/mickey/OICR-pytorch/lib/datasets/pascal_voc.py", line 137, in selectiv e_search_roidb roidb = self._load_selective_search_roidb(gt_roidb) File "/home/mickey/OICR-pytorch/lib/datasets/pascal_voc.py", line 154, in _load_se lective_search_roidb 'Selective search data not found at: {}'.format(filename) AssertionError: Selective search data not found at: /home/mickey/OICR-pytorch/data/s elective_search_data/voc_2007_trainval.mat

I'm a beginner, I don't understand a lot. Please help me. Thank you.

[jd730]

@zwyyy215 , The error above said that you do not have selective search files such as voc_2007_trainval.mat. I recommend you to follow the instruction of the original OICR repository this using this script

[jd730]

Or, you can directly download by using

wget https://dl.dropboxusercontent.com/s/orrt7o6bp6ae0tc/selective_search_data.tgz
tar -xvf selective_search_data.tgz

and move selective_search_data folder into data

Now, I added this instruction on README

[zwyyy215]

Thank you for your help. But I have new problems.I'm not sure how to modify it to make it work properly. Thank you again. (OICR) mickey@a208-System-Product-Name:~/OICR-pytorch$ CUDA_VISIBLE_DEVICES=0 python3 trainval_net.py Called with args: Namespace(batch_size=2, checkpoint=-1, checkpoint_interval=500, class_agnostic=False, cuda=False, dataset ='pascal_voc', disp_interval=50, large_scale=False, load_dir=None, lr=0.001, lr_decay_gamma=0.1, lr_decay _step=5, mGPUs=False, max_epochs=20, model='oicr', net='vgg16', num_workers=4, optimizer='sgd', save_dir= 'output', session=1, start_epoch=1, threshold=0.01, use_tb=False, vis=False) /home/mickey/OICR-pytorch/lib/model/utils/config.py:392: YAMLLoadWarning: calling yaml.load() without Loa der=... 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': [8, 16, 32], 'CROP_RESIZE_WITH_MAX_POOL': False, 'CUDA': False, 'DATA_DIR': '/home/mickey/OICR-pytorch/data', 'DEDUP_BOXES': 0.125, 'EPS': 1e-14, 'EXP_DIR': 'vgg16', 'FEAT_STRIDE': [16], 'GPU_ID': 0, 'MATLAB': 'matlab', 'MAX_NUM_GT_BOXES': 20, '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': 2, 'ROOT_DIR': '/home/mickey/OICR-pytorch', 'TEST': {'BBOX_REG': True, 'HAS_RPN': True, 'MAX_SIZE': 2000, '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': [480, 576, 688, 864, 1200], 'SVM': False, 'USE_FLIPPED': True}, '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': 2, 'LEARNING_RATE': 0.01, 'MAX_IMAGE_SIZE': 2000, 'MAX_ROIS_SIZE': 6000, 'MAX_SIZE': 2000, 'MOMENTUM': 0.9, 'PROPOSAL_LIMIT': 2000, 'PROPOSAL_METHOD': 'selective_search', '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': [480, 576, 688, 864, 1200], 'SNAPSHOT_ITERS': 5000, 'SNAPSHOT_KEPT': 3, 'SNAPSHOT_PREFIX': 'res101_faster_rcnn', 'STEPSIZE': [40000], 'SUMMARY_INTERVAL': 180, 'TRIM_HEIGHT': 600, 'TRIM_WIDTH': 600, 'TRUNCATED': False, 'USE_ALL_GT': True, 'USE_FLIPPED': True, 'USE_GT': False, 'USE_SEG': True, 'WEIGHT_DECAY': 0.0005}, 'USE_GPU_NMS': True} WARNING: You have a CUDA device, so you should probably run with --cuda selective_search Loaded dataset voc_2007_trainval for training Set proposal method: selective_search Appending horizontally-flipped training examples... voc_2007_trainval ss roidb loaded from /home/mickey/OICR-pytorch/data/cache/voc_2007_trainval_selective_s earch_roidb.pkl done Preparing training data... done before filtering, there are 10022 images... after filtering, there are 10022 images... voc_2007_trainval ss roidb loaded from /home/mickey/OICR-pytorch/data/cache/voc_2007_trainval_selective_s earch_roidb.pkl 5236 71 10022 roidb entries output/log/log.txt file already exists output/log/log_3.txt output/log/progress.csv file already exists Logging to output/log oicr Loading pretrained weights from data/pretrained_model/vgg16_caffe.pth Learning rate 0.001 Traceback (most recent call last): File "trainval_net.py", line 417, in rois, midn_loss, oc1, oc2, oc3, cls_prob= OICR(im_data, im_rois, labels, num_boxes) File "/home/mickey/anaconda3/envs/OICR/lib/python3.6/site-packages/torch/nn/modules/module.py", line 49 1, in call result = self.forward(*input, *kwargs) File "/home/mickey/OICR-pytorch/lib/model/oicr/oicr.py", line 139, in forward loss_oicr = WeightedSoftmaxWithLoss(self.ic_prob, label_ic, cls_loss_weights) File "/home/mickey/OICR-pytorch/lib/model/oicr/oicr.py", line 36, in WeightedSoftmaxWithLoss loss = (labels_ic torch.log(prob + eps)) RuntimeError: Expected object of type torch.cuda.FloatTensor but found type torch.FloatTensor for argumen t #2 'other'

[jd730]

@zwyyy215 This seems because cuda flag is off. could you run with --cuda?

Or run

CUDA_VISIBLE_DEVICES=0 python3 trainval_net.py --dataset pascal_voc --net vgg16 \
   --bs 2 --nw 6 --save_dir='results/test_single'  \
   --lr 0.001 --cuda --disp_interval 50 --vis \
        --checkpoint_interval=500 --model='oicr' --threshold=0.1

[zwyyy215]

Thank you. The program has finally run successfully. Thank you for your patient guidance.

[alexshaodong]

Thank you. The program has finally run successfully. Thank you for your patient guidance.

Have you solved this problem? Can you teach me?Thankyou！

[alexshaodong]

@zwyyy215 , The error above said that you do not have selective search files such as voc_2007_trainval.mat. I recommend you to follow the instruction of the original OICR repository this using this script

Have you solved this problem? Can you teach me?Thankyou！

[jd730]

@alexshaodong Did you confront the same problem?

As you cited, the problem of this issue came from not installing voc_2007_trainval.mat.

[alexshaodong]

@alexshaodong Did you confront the same problem?

As you cited, the problem of this issue came from not installing voc_2007_trainval.mat.

Hello, I came across this problem: Key Error:'boxes'.I have installed voc_2007_trainval.mat. I want to ask for your help! Thank you!

[alexshaodong]

Hi @zwyyy215, The gt flag is a legacy from faster-RCNN. The reason why you got a key error is, roidb of gt is composed of labels and seg (this).

For weakly supervised object detection, we should not use gt bounding boxes. However, if you really want to use gt boxes, you can simply change above code by using how annots.pkl is made, maybe changing parse_rec. (this).

Thank you.

This is the one. I want to ask you to teach me how to modify the code.

[jd730]

@alexshaodong Could you check whether voc_2007_trainval.mat is successfully loaded? check here

[jd730]

@alexshaodong If you want to use GT boxes, what you need to do is make new mat file. First, you need to parse GT boxes from annotations. In my implementation, annots.pkl is parsed boxes. So you can use annots.pkl to genetate mat file or, directly use the code below. this