My solution to FashionAI Key Points Detection of Apparel.
download data and put in data
data
├── r1-test-a
│ ├── Images
│ └── test.csv
├── r1-test-b
│ ├── Images
│ └── test.csv
├── r2-test-a
│ ├── Images
│ └── test.csv
├── r2-test-b
│ ├── Images
│ └── test.csv
└── train
├── Annotations
└── Images
There's a bug in mxnet/gluon/block.py
in SymbolBlock
. If you encounter save and load model issue, check a patch below.
class SymbolBlock(HybridBlock)
def __init__(self, outputs, inputs, params=None):
super(SymbolBlock, self).__init__(prefix=None, params=None)
self._prefix = ''
self._params = ParameterDict('', params)
self._reg_params = self._params # <--- add this line