test_nms_thresh and test_detections_per_img might be not defined in TridentNet

[tkm-n]

Hi, I want to evaluate trained TridentNet, I can train on my environment according to the instruction, but I cannot evaluate it. And, I found test_nms_thresh and test_detections_per_img might be not defined in "TridentRes5ROIHeads".

In detail, I ran the following command,

python train_net.py --config-file configs/tridentnet_fast_R_50_C4_1x.yaml --num-gpus 1 --eval-only MODEL.WEIGHTS output/model_0029999.pth
Then, I got this message,

[04/15 08:17:54 d2.data.common]: Serializing 5000 elements to byte tensors and concatenating them all ...
[04/15 08:17:54 d2.data.common]: Serialized dataset takes 19.39 MiB
[04/15 08:17:55 d2.evaluation.evaluator]: Start inference on 5000 images
Traceback (most recent call last):
  File "train_net.py", line 66, in <module>
    args=(args,),
  File "/detectron2/detectron2/engine/launch.py", line 52, in launch
    main_func(*args)
  File "train_net.py", line 49, in main
    res = Trainer.test(cfg, model)
  File "/detectron2/detectron2/engine/defaults.py", line 514, in test
    results_i = inference_on_dataset(model, data_loader, evaluator)
  File "/detectron2/detectron2/evaluation/evaluator.py", line 141, in inference_on_dataset
    outputs = model(inputs)
  File "/anaconda3/envs/detectron2/lib/python3.7/site-packages/torch/nn/modules/module.py", line 532, in __call__
    result = self.forward(*input, **kwargs)
  File "/detectron2/detectron2/modeling/meta_arch/rcnn.py", line 108, in forward
    return self.inference(batched_inputs)
  File "/detectron2/detectron2/modeling/meta_arch/rcnn.py", line 170, in inference
    results, _ = self.roi_heads(images, features, proposals, None)
  File "/anaconda3/envs/detectron2/lib/python3.7/site-packages/torch/nn/modules/module.py", line 532, in __call__
    result = self.forward(*input, **kwargs)
  File "/detectron2/projects/TridentNet/tridentnet/trident_rcnn.py", line 73, in forward
    pred_instances, num_branch, self.test_nms_thresh, self.test_detections_per_img
  File "/anaconda3/envs/detectron2/lib/python3.7/site-packages/torch/nn/modules/module.py", line 576, in __getattr__
    type(self).__name__, name))
AttributeError: 'TridentRes5ROIHeads' object has no attribute 'test_nms_thresh'

At that time, I used COCO dataset 2017 version, and I slightly changed the config file as below;
IMS_PER_BATCH: 3