v2.5.0 broke OD model training on custom dataset?

[mcdy143]

After updating to v2.5.0, training faster_rcnn model on customized dataset (COCO-like format) seem to have broken. The validation accuracy does stays close to 0 starting at the second epoch and stays 0. The trained model does not generate any bounding box even at confidence threshold=0.

After reverting back to v2.4.0, training worked like before using the same dataset and same config files. Validation output from one epoch: Evaluate annotation type *bbox* DONE (t=0.24s). Accumulating evaluation results... DONE (t=0.10s). Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.000 Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=1000 ] = 0.002 Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=1000 ] = 0.000 Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=1000 ] = -1.000 Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=1000 ] = 0.000 Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=1000 ] = 0.000 Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.002 Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=300 ] = 0.002 Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=1000 ] = 0.002 Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=1000 ] = -1.000 Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=1000 ] = 0.000 Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=1000 ] = 0.002

[RyanXLi]

Hi! Could you follow the template of reporting an error so we can get more information?

[mcdy143]

Sorry for the late reply. Please see below:

Reproduction

1. What command or script did you run?

   nohup python -u tools/train.py configs_od_entryway/cascade_rcnn/cascade_rcnn_x101_64x4d_fpn_1x_coco.py --work-dir work_dirs/OD_entryway_cascade_x101 &

2. Did you make any modifications on the code or config? Did you understand what you have modified? Yes. Modified files include: 1). Added custom CoCo format dataset under mmdetection/mmdet/datasets/ 2). Added new data class in mmdetection/mmdet/datasets/init.py 3). modified config files base/datasets/coco_detection.py and base/model/cascade_rcnn_r50_fpn.py 's num_classes to match number of classes in new dataset.

3. What dataset did you use? modified coco dataset with 12 classes

Environment

sys.platform: linux
Python: 3.7.8 | packaged by conda-forge | (default, Jul 31 2020, 02:25:08) [GCC 7.5.0]
CUDA available: True
GPU 0: Tesla V100-SXM2-16GB
CUDA_HOME: /usr/local/cuda
NVCC: Cuda compilation tools, release 10.0, V10.0.130
GCC: gcc (Ubuntu 7.5.0-3ubuntu1~18.04) 7.5.0
PyTorch: 1.5.0
PyTorch compiling details: PyTorch built with:
  - GCC 7.3
  - C++ Version: 201402
  - Intel(R) Math Kernel Library Version 2020.0.2 Product Build 20200624 for Intel(R) 64 architecture applications
  - Intel(R) MKL-DNN v0.21.1 (Git Hash 7d2fd500bc78936d1d648ca713b901012f470dbc)
  - OpenMP 201511 (a.k.a. OpenMP 4.5)
  - NNPACK is enabled
  - CPU capability usage: AVX2
  - CUDA Runtime 10.2
  - NVCC architecture flags: -gencode;arch=compute_37,code=sm_37;-gencode;arch=compute_50,code=sm_50;-gencode;arch=compute_60,code=sm_60;-gencode;arch=compute_61,code=sm_61;-gencode;arch=compute_70,code=sm_70;-gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_37,code=compute_37
  - CuDNN 7.6.5
  - Magma 2.5.2
  - Build settings: BLAS=MKL, BUILD_TYPE=Release, CXX_FLAGS= -Wno-deprecated -fvisibility-inlines-hidden -fopenmp -DNDEBUG -DUSE_FBGEMM -DUSE_QNNPACK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DUSE_INTERNAL_THREADPOOL_IMPL -O2 -fPIC -Wno-narrowing -Wall -Wextra -Werror=return-type -Wno-missing-field-initializers -Wno-type-limits -Wno-array-bounds -Wno-unknown-pragmas -Wno-sign-compare -Wno-unused-parameter -Wno-unused-variable -Wno-unused-function -Wno-unused-result -Wno-strict-overflow -Wno-strict-aliasing -Wno-error=deprecated-declarations -Wno-stringop-overflow -Wno-error=pedantic -Wno-error=redundant-decls -Wno-error=old-style-cast -fdiagnostics-color=always -faligned-new -Wno-unused-but-set-variable -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Wno-stringop-overflow, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, PERF_WITH_AVX512=1, USE_CUDA=ON, USE_EXCEPTION_PTR=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_MKL=ON, USE_MKLDNN=ON, USE_MPI=OFF, USE_NCCL=ON, USE_NNPACK=ON, USE_OPENMP=ON, USE_STATIC_DISPATCH=OFF, 

TorchVision: 0.6.0a0+82fd1c8
OpenCV: 4.4.0
MMCV: 1.2.0
MMCV Compiler: GCC 7.3
MMCV CUDA Compiler: 10.2
MMDetection: 2.6.0+885b86b

Error traceback If applicable, paste the error trackback here.

Evaluate annotation type *bbox* DONE (t=0.24s). Accumulating evaluation results... DONE (t=0.10s). Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.000 Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=1000 ] = 0.002 Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=1000 ] = 0.000 Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=1000 ] = -1.000 Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=1000 ] = 0.000 Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=1000 ] = 0.000 Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.002 Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=300 ] = 0.002 Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=1000 ] = 0.002 Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=1000 ] = -1.000 Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=1000 ] = 0.000 Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=1000 ] = 0.002

[ZwwWayne]

This is because that the version before MMDetection v2.6 has issues with the customized classes on COCODataset. Please try the newest MMDet and see whether you still meet the issue.