-
det from_scratch seems fine, with 36.4 AP
but fine_tune only has 19.6 AP
the only thing I modified is reusing the MSRA weight
> WEIGHT: "catalog://ImageNetPretrained/MSRA/R-50"
and register "R…
-
command used :
`python tools/run_net.py --cfg configs/AVA/C2D_8x8_R50_SHORT.yaml DATA.PATH_TO_DATA_DIR ../AVA\ Actions\ Main/data/ava/ NUM_GPUS 2 TRAIN.BATCH_SIZE 16 AVA.FRAME_LIST_DIR ../AVA\ Action…
-
Hi, thanks for your great codebase. When I run
python tools/run_net.py --cfg configs/AVA/SLOWFAST_32x2_R50_SHORT.yaml NUM_GPUS 2
it appear the following error.. it looks like a compile errro abo…
-
when i python tools/train.py configs/pn_test.py, pn_test.py is similar to mask-r-cnn.
2020-03-08 16:46:36,861 - mmdet - INFO - Environment info:
---------------------------------------------------…
-
For R-101 Faster R-CNN, baseline ap is 0.540 , Iou loss is 0.32 ,giou loss is 0.43. How to solve it.
# model settings
model = dict(
type='FasterRCNN',
pretrained='torchvision://resnet101',…
-
Going through the code, I see there is already support for ROI Pooling CUDA kernels. It would be great to have support for various ROI Pooling operations as easy to add layers to facilitate research i…
-
**Describe the bug**
The training speed of Faster-R-CNN-Caffe-c4 on pascal VOC 2007 is too slow.
Mmdetection log:
```
2020-05-08 18:32:34,938 - mmdet - INFO - Start running, host: aistudio@j…
-
Hello,
I found that the 'anchor_scale' is being used in faster_rcnn config files, and have tried to see what it does exactly.
Whenever I tried to change this from [8] to some other values or multi…
-
I use mask-rcnn benchmark framework, and replace the backbone with resnest. But loss is always nan.
```
...
Gradient overflow. Skipping step, loss scaler 0 reducing loss scale to 2.06795153138256…
-
**Describe the bug**
when using ghm loss,error occured.
**Reproduction**
1. script
```
fp16 = dict(loss_scale=512.)
# model settings
model = dict(
type='CascadeRCNN',
num_stages…