search

leoxiaobin / deep-high-resolution-net.pytorch

The project is an official implementation of our CVPR2019 paper "Deep High-Resolution Representation Learning for Human Pose Estimation"
https://jingdongwang2017.github.io/Projects/HRNet/PoseEstimation.html
MIT License
4.31k stars 908 forks source link

RuntimeError: CUDA error: out of memory -- although GPU is empty #288

Open Masum06 opened 2 years ago

Masum06 commented 2 years ago

Only reason I can think of is my cuda version is 11.7, but the latest version of PyTorch available is for cuda 11.6. Could that be the issue?

image

Log:

 python tools/test.py \
    --cfg experiments/coco/hrnet/w48_384x288_adam_lr1e-3.yaml \
    TEST.MODEL_FILE models/pytorch/pose_hrnet_w48_384x288.pth \
    TEST.USE_GT_BBOX False

=> creating log/coco/pose_hrnet/w48_384x288_adam_lr1e-3_2022-07-19-07-56
Namespace(cfg='experiments/coco/hrnet/w48_384x288_adam_lr1e-3.yaml', opts=['TEST.MODEL_FILE', 'models/pytorch/pose_hrnet_w48_384x288.pth', 'TEST.USE_GT_BBOX', 'False'], modelDir='', logDir='', dataDir='', prevModelDir='')
AUTO_RESUME: True
CUDNN:
  BENCHMARK: True
  DETERMINISTIC: False
  ENABLED: True
DATASET:
  COLOR_RGB: True
  DATASET: coco
  DATA_FORMAT: jpg
  FLIP: True
  HYBRID_JOINTS_TYPE: 
  NUM_JOINTS_HALF_BODY: 8
  PROB_HALF_BODY: 0.3
  ROOT: data/coco/
  ROT_FACTOR: 45
  SCALE_FACTOR: 0.35
  SELECT_DATA: False
  TEST_SET: val2017
  TRAIN_SET: train2017
DATA_DIR: 
DEBUG:
  DEBUG: True
  SAVE_BATCH_IMAGES_GT: True
  SAVE_BATCH_IMAGES_PRED: True
  SAVE_HEATMAPS_GT: True
  SAVE_HEATMAPS_PRED: True
GPUS: (0, 1, 2, 3)
LOG_DIR: log
LOSS:
  TOPK: 8
  USE_DIFFERENT_JOINTS_WEIGHT: False
  USE_OHKM: False
  USE_TARGET_WEIGHT: True
MODEL:
  EXTRA:
    FINAL_CONV_KERNEL: 1
    PRETRAINED_LAYERS: ['conv1', 'bn1', 'conv2', 'bn2', 'layer1', 'transition1', 'stage2', 'transition2', 'stage3', 'transition3', 'stage4']
    STAGE2:
      BLOCK: BASIC
      FUSE_METHOD: SUM
      NUM_BLOCKS: [4, 4]
      NUM_BRANCHES: 2
      NUM_CHANNELS: [48, 96]
      NUM_MODULES: 1
    STAGE3:
      BLOCK: BASIC
      FUSE_METHOD: SUM
      NUM_BLOCKS: [4, 4, 4]
      NUM_BRANCHES: 3
      NUM_CHANNELS: [48, 96, 192]
      NUM_MODULES: 4
    STAGE4:
      BLOCK: BASIC
      FUSE_METHOD: SUM
      NUM_BLOCKS: [4, 4, 4, 4]
      NUM_BRANCHES: 4
      NUM_CHANNELS: [48, 96, 192, 384]
      NUM_MODULES: 3
  HEATMAP_SIZE: [72, 96]
  IMAGE_SIZE: [288, 384]
  INIT_WEIGHTS: True
  NAME: pose_hrnet
  NUM_JOINTS: 17
  PRETRAINED: models/pytorch/imagenet/hrnet_w48-8ef0771d.pth
  SIGMA: 3
  TAG_PER_JOINT: True
  TARGET_TYPE: gaussian
OUTPUT_DIR: output
PIN_MEMORY: True
PRINT_FREQ: 100
RANK: 0
TEST:
  BATCH_SIZE_PER_GPU: 24
  BBOX_THRE: 1.0
  COCO_BBOX_FILE: data/coco/person_detection_results/COCO_val2017_detections_AP_H_56_person.json
  FLIP_TEST: True
  IMAGE_THRE: 0.0
  IN_VIS_THRE: 0.2
  MODEL_FILE: models/pytorch/pose_hrnet_w48_384x288.pth
  NMS_THRE: 1.0
  OKS_THRE: 0.9
  POST_PROCESS: True
  SHIFT_HEATMAP: True
  SOFT_NMS: False
  USE_GT_BBOX: False
TRAIN:
  BATCH_SIZE_PER_GPU: 24
  BEGIN_EPOCH: 0
  CHECKPOINT: 
  END_EPOCH: 210
  GAMMA1: 0.99
  GAMMA2: 0.0
  LR: 0.001
  LR_FACTOR: 0.1
  LR_STEP: [170, 200]
  MOMENTUM: 0.9
  NESTEROV: False
  OPTIMIZER: adam
  RESUME: False
  SHUFFLE: True
  WD: 0.0001
WORKERS: 24
=> loading model from models/pytorch/pose_hrnet_w48_384x288.pth
loading annotations into memory...
Done (t=0.12s)
creating index...
index created!
=> classes: ['__background__', 'person']
=> num_images: 5000
=> Total boxes: 104125
=> Total boxes after fliter low score@0.0: 104125
=> load 104125 samples
Traceback (most recent call last):
  File "/mnt/e/hi_5/deep-high-resolution-net.pytorch/tools/test.py", line 130, in <module>
    main()
  File "/mnt/e/hi_5/deep-high-resolution-net.pytorch/tools/test.py", line 125, in main
    validate(cfg, valid_loader, valid_dataset, model, criterion,
  File "/mnt/e/hi_5/deep-high-resolution-net.pytorch/tools/../lib/core/function.py", line 118, in validate
    for i, (input, target, target_weight, meta) in enumerate(val_loader):
  File "/home/roc-hci/anaconda3/envs/hi5/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 652, in __next__
    data = self._next_data()
  File "/home/roc-hci/anaconda3/envs/hi5/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 1347, in _next_data
    return self._process_data(data)
  File "/home/roc-hci/anaconda3/envs/hi5/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 1373, in _process_data
    data.reraise()
  File "/home/roc-hci/anaconda3/envs/hi5/lib/python3.10/site-packages/torch/_utils.py", line 461, in reraise
    raise exception
RuntimeError: Caught RuntimeError in pin memory thread for device 0.
Original Traceback (most recent call last):
  File "/home/roc-hci/anaconda3/envs/hi5/lib/python3.10/site-packages/torch/utils/data/_utils/pin_memory.py", line 34, in _pin_memory_loop
    data = pin_memory(data, device)
  File "/home/roc-hci/anaconda3/envs/hi5/lib/python3.10/site-packages/torch/utils/data/_utils/pin_memory.py", line 65, in pin_memory
    return type(data)([pin_memory(sample, device) for sample in data])  # type: ignore[call-arg]
  File "/home/roc-hci/anaconda3/envs/hi5/lib/python3.10/site-packages/torch/utils/data/_utils/pin_memory.py", line 65, in <listcomp>
    return type(data)([pin_memory(sample, device) for sample in data])  # type: ignore[call-arg]
  File "/home/roc-hci/anaconda3/envs/hi5/lib/python3.10/site-packages/torch/utils/data/_utils/pin_memory.py", line 50, in pin_memory
    return data.pin_memory(device)
RuntimeError: CUDA error: out of memory
CUDA kernel errors might be asynchronously reported at some other API call,so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1.

terminate called without an active exception```

### --------------------------------

> nvidia-smi
```Tue Jul 19 07:54:49 2022       
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 515.57       Driver Version: 516.59       CUDA Version: 11.7     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|                               |                      |               MIG M. |
|===============================+======================+======================|
|   0  NVIDIA GeForce ...  On   | 00000000:0A:00.0  On |                  N/A |
|  0%   49C    P8    24W / 370W |    683MiB / 24576MiB |      2%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+
|   1  NVIDIA GeForce ...  On   | 00000000:0B:00.0 Off |                  N/A |
|  0%   39C    P8    14W / 370W |      0MiB / 24576MiB |      0%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes:                                                                  |
|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |
|        ID   ID                                                   Usage      |
|=============================================================================|
|  No running processes found                                                 |
+-----------------------------------------------------------------------------+```