Merge pull request #74 from ucbdrive/refactor

[Refactor] Fixing refactor

.gitignore

@@ -21,6 +21,7 @@ checkpoints/
 
 # datasets
 datasets/coco
+datasets/cocosplit
 datasets/VOC2007
 datasets/VOC2012
 datasets/vocsplit

README.md

@@ -53,6 +53,10 @@ python3 -m venv fsdet
 source fsdet/bin/activate
 ```
 You can also use `conda` to create a new environment.
+```angular2html
+conda create --name fsdet
+conda activate fsdet
+```
 * Install Pytorch 1.6 with CUDA 10.2 
 ```angular2html
 pip install torch torchvision
@@ -108,7 +112,7 @@ We provide a set of benchmark results and pre-trained models available for downl
   for example, `COCO-detection/faster_rcnn_R_101_FPN_ft_all_1shot.yaml`.
 2. We provide `demo.py` that is able to run builtin standard models. Run it with:
 ```
-python demo/demo.py --config-file configs/COCO-detection/faster_rcnn_R_101_FPN_ft_all_1shot.yaml \
+python3 -m demo.demo --config-file configs/COCO-detection/faster_rcnn_R_101_FPN_ft_all_1shot.yaml \
   --input input1.jpg input2.jpg \
   [--other-options]
   --opts MODEL.WEIGHTS fsdet://coco/tfa_cos_1shot/model_final.pth
@@ -146,13 +150,12 @@ For ease of training and evaluation over multiple runs, we provided several help
 
 You can use `tools/run_experiments.py` to do the training and evaluation. For example, to experiment on 30 seeds of the first split of PascalVOC on all shots, run
 ```angular2html
-python tools/run_experiments.py --num-gpus 8 \
+python3 -m tools.run_experiments --num-gpus 8 \
         --shots 1 2 3 5 10 --seeds 0 30 --split 1
 ```
 
 After training and evaluation, you can use `tools/aggregate_seeds.py` to aggregate the results over all the seeds to obtain one set of numbers. To aggregate the 3-shot results of the above command, run
 ```angular2html
-python tools/aggregate_seeds.py --shots 3 --seeds 30 --split 1 \
+python3 -m tools.aggregate_seeds --shots 3 --seeds 30 --split 1 \
         --print --plot
 ```
-

demo/demo.py

@@ -6,10 +6,11 @@
 import multiprocessing as mp
 import os
 import time
-from detectron2.config import get_cfg
+
+from demo.predictor import VisualizationDemo
 from detectron2.data.detection_utils import read_image
 from detectron2.utils.logger import setup_logger
-from predictor import VisualizationDemo
+from fsdet.config import get_cfg
 
 # constants
 WINDOW_NAME = "COCO detections"

demo/predictor.py

@@ -6,9 +6,9 @@
 import multiprocessing as mp
 from collections import deque
 from detectron2.data import MetadataCatalog
-from detectron2.engine.defaults import DefaultPredictor
 from detectron2.utils.video_visualizer import VideoVisualizer
 from detectron2.utils.visualizer import ColorMode, Visualizer
+from fsdet.engine import DefaultPredictor
 
 
 class VisualizationDemo(object):

docs/TRAIN_INST.md

@@ -8,7 +8,7 @@ TFA is trained in two stages. We first train the entire object detector on the d
 
 First train a base model. To train a base model on the first split of PASCAL VOC, run
 ```angular2html
-python tools/train_net.py --num-gpus 8 \
+python3 -m tools.train_net --num-gpus 8 \
         --config-file configs/PascalVOC-detection/split1/faster_rcnn_R_101_base1.yaml
 ```
 
@@ -22,7 +22,7 @@ After training the base model, run ```tools/ckpt_surgery.py``` to obtain an init
 
 To randomly initialize the weights corresponding to the novel classes, run
 ```angular2html
-python tools/ckpt_surgery.py \
+python3 -m tools.ckpt_surgery \
         --src1 checkpoints/voc/faster_rcnn/faster_rcnn_R_101_FPN_base1/model_final.pth \
         --method randinit \
         --save-dir checkpoints/voc/faster_rcnn/faster_rcnn_R_101_FPN_all1
@@ -33,22 +33,22 @@ The resulting weights will be saved to `checkpoints/voc/faster_rcnn/faster_rcnn_
 
 To use novel weights, fine-tune a predictor on the novel set. We reuse the base model trained in the previous stage but retrain the last layer from scratch. First remove the last layer from the weights file by running
 ```angular2html
-python tools/ckpt_surgery.py \
+python3 -m tools.ckpt_surgery \
         --src1 checkpoints/voc/faster_rcnn/faster_rcnn_R_101_FPN_base1/model_final.pth \
         --method remove \
         --save-dir checkpoints/voc/faster_rcnn/faster_rcnn_R_101_FPN_all1
 ```
 
 Next, fine-tune the predictor on the novel set by running
 ```angular2html
-python tools/train_net.py --num-gpus 8 \
+python3 -m tools.train_net --num-gpus 8 \
         --config-file configs/PascalVOC-detection/split1/faster_rcnn_R_101_FPN_ft_novel1_1shot.yaml \
         --opts MODEL.WEIGHTS checkpoints/voc/faster_rcnn/faster_rcnn_R_101_FPN_all1/model_reset_remove.pth
 ```
 
 Finally, combine the base weights from the base model with the novel weights by running
 ```angular2html
-python tools/ckpt_surgery.py \
+python3 -m tools.ckpt_surgery \
         --src1 checkpoints/voc/faster_rcnn/faster_rcnn_R_101_FPN_base1/model_final.pth \
         --src2 checkpoints/voc/faster_rcnn/faster_rcnn_R_101_FPN_ft_novel1_1shot/model_final.pth \
         --method combine \
@@ -60,7 +60,7 @@ The resulting weights will be saved to `checkpoints/voc/faster_rcnn/faster_rcnn_
 
 We will then fine-tune the last layer of the full model on a balanced dataset by running
 ```angular2html
-python tools/train_net.py --num-gpus 8 \
+python3 -m tools.train_net --num-gpus 8 \
         --config-file configs/PascalVOC-detection/split1/faster_rcnn_R_101_FPN_ft_all1_1shot.yaml \
         --opts MODEL.WEIGHTS WEIGHTS_PATH
 ```

fsdet/checkpoint/__init__.py

@@ -1 +1,4 @@
 from . import catalog as _UNUSED  # register the handler
+from .detection_checkpoint import DetectionCheckpointer
+
+__all__ = ["DetectionCheckpointer"]