The reward goes down

README.md

@@ -7,6 +7,12 @@ This repo contains the Pytorch implementation of the AAAI'18 paper - [Deep Reinf
 
 The main requirements are [pytorch](http://pytorch.org/) (`v0.4.0`) and python `2.7`. Some dependencies that may not be installed in your machine are [tabulate](https://pypi.org/project/tabulate/) and [h5py](https://github.com/h5py/h5py). Please install other missing dependencies.
 
+## TODO
+1. Edit README
+2. Reconstruct Repository
+3. ~~Add KTS(Kernel Temporal Segmentation)~~
+4. Edit code for test
+
 ## Get started
 1. Download preprocessed datasets
 ```bash
@@ -16,6 +22,11 @@ cd pytorch-vsumm-reinforce
 wget http://www.eecs.qmul.ac.uk/~kz303/vsumm-reinforce/datasets.tar.gz
 tar -xvzf datasets.tar.gz
 ```
+* If can't download, open following link.(I loaded same dataset in my onedrive)
+```
+https://onedrive.live.com/?authkey=%21AO1tsqjDVCeakGg&cid=6FD3437627D709EE&id=6FD3437627D709EE%212809&parId=root&action=locate
+```
+
 2. Make splits
 ```bash
 python create_split.py -d datasets/eccv16_dataset_summe_google_pool5.h5 --save-dir datasets --save-name summe_splits  --num-splits 5
@@ -25,13 +36,23 @@ As a result, the dataset is randomly split for 5 times, which are saved as json
 Train and test codes are written in `main.py`. To see the detailed arguments, please do `python main.py -h`.
 
 ## How to train
+* Edit config/config.py ( see config/README.txt )
+
 ```bash
-python main.py -d datasets/eccv16_dataset_summe_google_pool5.h5 -s datasets/summe_splits.json -m summe --gpu 0 --save-dir log/summe-split0 --split-id 0 --verbose
+python video_summarization.py
+```
+## How to evaluate
+* Edit config/config.py ( see config/README.txt )
+
+```bash
+python video_summarization.py
 ```
 
 ## How to test
+* Edit config/config.py ( see config/README.txt )
+
 ```bash
-python main.py -d datasets/eccv16_dataset_summe_google_pool5.h5 -s datasets/summe_splits.json -m summe --gpu 0 --save-dir log/summe-split0 --split-id 0 --evaluate --resume path_to_your_model.pth.tar --verbose --save-results
+python video_summarization.py
 ```
 
 If argument `--save-results` is enabled, output results will be saved to `results.h5` under the same folder specified by `--save-dir`. To visualize the score-vs-gtscore, simple do
@@ -73,6 +94,7 @@ Please remember to specify the naming format of your video frames on this [line]
 ## How to use your own data
 We preprocess data by extracting image features for videos and save them to `h5` file. The file format looks like [this](https://github.com/KaiyangZhou/vsumm-reinforce/issues/1#issuecomment-363492711). After that, you can make split via `create_split.py`. If you wanna train policy network using the entire dataset, just do `train_keys = dataset.keys()`. [Here](https://github.com/KaiyangZhou/pytorch-vsumm-reinforce/blob/master/main.py#L75) is the code where we initialize dataset. If you have any problems, feel free to contact me by email or raise an `issue`.
 
+
 ## Citation
 ```
 @article{zhou2017reinforcevsumm, 
@@ -81,4 +103,4 @@ We preprocess data by extracting image features for videos and save them to `h5`
    journal={arXiv:1801.00054}, 
    year={2017} 
 }
-```
+```

config/README.txt

@@ -0,0 +1,34 @@
+Configuration README
+====================
+
+1. example config to train
+# Dataset options
+DATASET = 'datasets/eccv16_dataset_summe_google_pool5.h5'
+SPLIT = 'datasets/summe_splits.json'
+SPLIT_ID = 0
+METRIC = 'summe'
+
+# Misc
+GPU = '0'
+EVALUATE = False
+TEST = False
+VERBOSE = True
+SAVE_DIR = 'log/summe-split0'
+
+2. example config to evaluate
+# Dataset options
+DATASET = 'datasets/eccv16_dataset_summe_google_pool5.h5'
+SPLIT = 'datasets/summe_splits.json'
+SPLIT_ID = 0
+METRIC = 'summe'
+
+# Misc
+GPU = '0'
+EVALUATE = True
+TEST = False
+RESUME = 'log/summe-split0/model_epoch60.pth.tar'
+VERBOSE = True
+SAVE_DIR = 'log/summe-split0'
+SAVE_RESULT = True
+
+3. example config to test

config/__init__.py

@@ -0,0 +1 @@
+from config import *

config/config.py

@@ -0,0 +1,32 @@
+# ============ TRAIN CONFIG ==============
+# Dataset options
+DATASET = 'datasets/eccv16_dataset_summe_google_pool5.h5' # path to h5 dataset (required)
+SPLIT = 'datasets/summe_splits.json' # path to split file (required)
+SPLIT_ID = 0 # split index (default: 0)
+METRIC = 'summe' # evaluation metric ['tvsum', 'summe'])
+
+# Model options
+INPUT_DIM = 1024 # input dimension (default: 1024)
+HIDDEN_DIM = 256 # hidden unit dimension of DSN (default: 256)
+NUM_LAYERS = 1 # number of RNN layers (default: 1)
+RNN_CELL = 'lstm' # RNN cell type (default: lstm)
+
+# Optimization options
+LR = 1e-05 # learning rate (default: 1e-05)
+WEIGHT_DECAY = 1e-05 # weight decay rate (default: 1e-05)
+MAX_EPOCH = 60 # maximum epoch for training (default: 60)
+STEP_SIZE = 30 # how many steps to decay learning rate (default: 30)
+GAMMA = 0.1 # learning rate decay (default: 0.1)
+NUM_EPISODE = 5 # number of episodes (default: 5)
+BETA = 0.01 # weight for summary length penalty term (default: 0.01)
+
+# Misc
+SEED = 1 # random seed (default: 1)
+GPU = '0' # which gpu devices to use (default: 0)
+USE_CPU = False # use cpu device
+EVALUATE = False # whether to do evaluation only
+TEST = False # whether to do evaluation only
+RESUME = '' # path to resume file
+VERBOSE = True # whether to show detailed test results
+SAVE_DIR = 'log/summe-split0' # path to save output (default: log/)
+SAVE_RESULTS = True # whether to save output results

create_split.py

@@ -1,26 +1,29 @@
+# Dataset Split
+
 from __future__ import print_function
+
 import os
-import os.path as osp
 import argparse
 import h5py
 import math
 import numpy as np
 
-from utils import write_json
+from utils.file_process import write_json
 
 parser = argparse.ArgumentParser("Code to create splits in json form")
-parser.add_argument('-d', '--dataset', type=str, required=True, help="path to h5 dataset (required)")
-parser.add_argument('--save-dir', type=str, default='datasets', help="path to save output json file (default: 'datasets/')")
-parser.add_argument('--save-name', type=str, default='splits', help="name to save as, excluding extension (default: 'splits')")
-parser.add_argument('--num-splits', type=int, default=5, help="how many splits to generate (default: 5)")
-parser.add_argument('--train-percent', type=float, default=0.8, help="percentage of training data (default: 0.8)")
+parser.add_argument("-d", "--dataset", type=str, required=True, help="path to h5 dataset (required)")
+parser.add_argument("--save-dir", type=str, default='datasets', help="path to save output jon file (default: 'datasets/'")
+parser.add_argument("--save-name", type=str, default="splits", help="name to save as, excluding extension (default: 'splits')")
+parser.add_argument("--num-splits", type=int, default=5, help="how many splits to generate (default: 5)")
+parser.add_argument("--train-percent", type=float, default=0.8, help="percentage of training data (default: 0.8)")
 
 args = parser.parse_args()
 
 def split_random(keys, num_videos, num_train):
-    """Random split"""
+    """ Random split """
     train_keys, test_keys = [], []
     rnd_idxs = np.random.choice(range(num_videos), size=num_train, replace=False)
+
     for key_idx, key in enumerate(keys):
         if key_idx in rnd_idxs:
             train_keys.append(key)
@@ -32,29 +35,32 @@ def split_random(keys, num_videos, num_train):
     return train_keys, test_keys
 
 def create():
-    print("==========\nArgs:{}\n==========".format(args))
+    print("===========\nArgs:{}\n=========".format(args))
     print("Goal: randomly split data for {} times, {:.1%} for training and the rest for testing".format(args.num_splits, args.train_percent))
-    print("Loading dataset from {}".format(args.dataset))
+    print("Loading dataset from: {}".format(args.dataset))
+
     dataset = h5py.File(args.dataset, 'r')
     keys = dataset.keys()
     num_videos = len(keys)
     num_train = int(math.ceil(num_videos * args.train_percent))
     num_test = num_videos - num_train
-    print("Split breakdown: # total videos {}. # train videos {}. # test videos {}".format(num_videos, num_train, num_test))
+    print("Split breakdown: # total videos {}. # train videos. # test videos {}".format(num_videos, num_train, num_test))
     splits = []
 
     for split_idx in range(args.num_splits):
         train_keys, test_keys = split_random(keys, num_videos, num_train)
+
         splits.append({
             'train_keys': train_keys,
             'test_keys': test_keys,
-            })
+        })
 
-    saveto = osp.join(args.save_dir, args.save_name + '.json')
-    write_json(splits, saveto)
-    print("Splits saved to {}".format(saveto))
+    save_path = os.path.join(args.save_dir, args.save_name + '.json')
+    write_json(splits, save_path)
+    print("Splits save to {}".format(save_path))
 
     dataset.close()
 
 if __name__ == '__main__':
-    create()
+    create()
+