def _setup_graph(self):
self.pred = self.trainer.get_predictor(
['image'], ['fastrcnn_fg_probs', 'fastrcnn_fg_boxes'])
self.df = PrefetchDataZMQ(get_eval_dataflow(), 1)
EVAL_TIMES = 5 # eval 5 times during training
interval = self.trainer.config.max_epoch // (EVAL_TIMES + 1)
self.epochs_to_eval = set([interval * k for k in range(1, EVAL_TIMES)])
self.epochs_to_eval.add(self.trainer.config.max_epoch)
get_tf_nms() # just to make sure the nms part of graph is created
def _setup_graph(self):
self.pred = self.trainer.get_predictor(
['image'], ['fastrcnn_fg_probs', 'fastrcnn_fg_boxes'])
self.df = PrefetchDataZMQ(get_eval_dataflow(), 1)
get_tf_nms() # just to make sure the nms part of graph is created