https://github.com/tonysy/mx-rcnn-flow
https://github.com/tonysy/mx-rcnn-flow
Last synced: about 2 months ago
JSON representation
- Host: GitHub
- URL: https://github.com/tonysy/mx-rcnn-flow
- Owner: tonysy
- License: other
- Created: 2017-04-12T08:16:00.000Z (about 8 years ago)
- Default Branch: master
- Last Pushed: 2017-06-06T06:38:50.000Z (about 8 years ago)
- Last Synced: 2025-02-14T11:52:59.900Z (4 months ago)
- Language: Python
- Size: 297 KB
- Stars: 2
- Watchers: 3
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Deep Feature R-CNN
## Deep Feature flow
modified the orginal Faster R-CNN mxnet version.
## Results Record
### 1. RCNN
#### Results
kitti | easy | medium | hard
:---: | :---: | :----: | :---:
car | 0.8782325455 | 0.7572223636 | 0.6334450909
pedestrian | 0.7038934545 | 0.6196088182 | 0.5497168182
cyclist | 0.7561599091 | 0.5126151818 | 0.4845294545
#### Methods
- optimizer: sgd
#### Training Logs
```
04-05 01:28 root INFO Epoch[9] Train-RPNAcc=0.998872
04-05 01:28 root INFO Epoch[9] Train-RPNLogLoss=0.003826
04-05 01:28 root INFO Epoch[9] Train-RPNL1Loss=0.118832
04-05 01:28 root INFO Epoch[9] Train-RCNNAcc=0.980337
04-05 01:28 root INFO Epoch[9] Train-RCNNLogLoss=0.051186
04-05 01:28 root INFO Epoch[9] Train-RCNNL1Loss=0.653532
04-05 01:28 root INFO Epoch[9] Time cost=1299.938
04-05 01:28 root INFO Saved checkpoint to "model/e2e-rcnn-log-0010.params"
```
### 2. RCNN-DFF-fix-flow
#### Method:
- Use `mx.symbol.GridGenerator` and `mx.symbol.BilinearSampler`
- `config.FLOW_SCALE_FACTOR = 0.00390625`
#### Results
kitti | easy | medium | hard
:---: | :---: | :----: | :---:
car | 0.707715727273 | 0.545954545455 | 0.470983818182
pedestrian | 0.495135727273 | 0.414654636364 | 0.337167090909
cyclist | 0.504706545455 | 0.310941636364 | 0.304307272727
### 3. RCNN-DFF-non-fix-flow
#### Method:
- Use `mx.symbol.GridGenerator` and `mx.symbol.BilinearSampler`
- `config.FLOW_SCALE_FACTOR = 0.00390625`
#### Results
1. Epoch 8
kitti | easy | medium | hard
:---: | :---: | :----: | :---:
car | 0.793966090909 | 0.637243 | 0.553690181818
pedestrian | 0.504789090909 | 0.419259909091 | 0.402404181818
cyclist | 0.529568454545 | 0.317203 | 0.313888272727
1. Epoch 10
kitti | easy | medium | hard
:---: | :---: | :----: | :---:
car | 0.796542363636 | 0.638948727273 | 0.555344363636
pedestrian | 0.496717818182 | 0.414531636364 | 0.401392
cyclist | 0.532543363636 | 0.317411909091 | 0.316632636364
## Structure
This repository provides Faster R-CNN as a package named `rcnn`.
* `rcnn.core`: core routines in Faster R-CNN training and testing.
* `rcnn.cython`: cython speedup from py-faster-rcnn.
* `rcnn.dataset`: dataset library. Base class is `rcnn.dataset.imdb.IMDB`.
* `rcnn.io`: prepare training data.
* `rcnn.processing`: data and label processing library.
* `rcnn.symbol`: symbol and operator.
* `rcnn.test`: test utilities.
* `rcnn.tools`: training and testing wrapper.
* `rcnn.utils`: utilities in training and testing, usually overloads mxnet functions.