add readme

datasets/get_dataset_cuhk03np.sh

@@ -1,5 +1,3 @@
-echo 'Please enter your user name for bitbucket'
-read username
-wget --user=$username --ask-password https://bitbucket.org/eric_thesis/reid2/downloads/cuhk03-np.zip
+wget https://drive.google.com/uc?export=download&confirm=g71d&id=1pBCIAGSZ81pgvqjC-lUHtl0OYV1icgkz
 unzip cuhk03-np.zip
 rm -rf cuhk03-np.zip

datasets/get_dataset_dukemtmc.sh

@@ -0,0 +1,6 @@
+echo 'Please enter your user name for bitbucket'
+read username
+wget https://drive.google.com/uc?id=0B0VOCNYh8HeRdnBPa2ZWaVBYSVk
+mv DukeMTMC-reID.zip dukemtmc-reid.zip
+unzip dukemtmc-reid.zip
+rm -rf dukemtmc-reid.zip

datasets/get_dataset_market1501.sh

@@ -1,5 +1,6 @@
 echo 'Please enter your user name for bitbucket'
 read username
-wget --user=$username --ask-password https://bitbucket.org/eric_thesis/reid2/downloads/market1501.zip
+wget https://drive.google.com/file/d/0B8-rUzbwVRk0c054eEozWG9COHM/view
+mv Market-1501-v15.09.15.zip market1501.zip
 unzip market1501.zip
 rm -rf market1501.zip

readme.md

@@ -0,0 +1,43 @@
+# Person Re-Identification Baseline
+
+This repo basically is the re-implementation (tensorflow + keras) of a person
+re-identification baseline propose by the paper
+["Bag of Tricks and A Strong Baseline for Deep Person Re-identification"](https://arxiv.org/abs/1903.07071).
+
+The authors original implementation which is in Pytorch can be found in their
+[repo](https://github.com/michuanhaohao/reid-strong-baseline).
+
+
+## Enviornment
+
+A powerful GPU is required for running the code, with Nivida GTX 1070ti, a training with the standard 120 epochs
+will take almost 4 hours.
+
+This implementatoin is based on tensorflow and keras (currently not other backend are suppoted rather
+than `tf`), the tested version are listed below:
+
+- python:               3.6.7
+- tensorflow:           1.12.0
+- tensorflow-base:      1.12.0
+- tensorflow-gpu:       1.12.0
+- keras:                2.2.4
+- keras-applications:   1.0.6
+- keras-base:           2.2.4
+- keras-preprocessing:  1.0.5
+
+## Run
+
+Before you run, you need to speicify the dataset directory in your local machine. Go to the `reid.py` file,
+check the global variable named `g_data_root`. If you don't have the dataset yet, you can get the dataset by
+using the srcipt in the `datasets` folder. If you do so, set `g_data_root = './datasets'`.
+
+To train or try the model out, go to the very end of the `reid.py` file. Comment the method you don't want
+and uncomment the method you want then launch the terminal and run:
+
+```
+python reid.py
+```
+
+## Theory
+
+For the theory behind the code, please check with the wiki.

reid.py

@@ -27,6 +27,7 @@
 
 ''' global variables '''
 g_data_root  = '/home/h_lai/Documents/dl/reid/triplet/datasets'
+# g_data_root = './datasets'
 g_output_dir = './output'
 
 g_num_ids  = 16
@@ -232,6 +233,20 @@ def train_and_test(weight_path):
     e = Evaluator(dataset, feat_model, g_img_h, g_img_w)
     e.compute()
 
+def train(target_model, save_weight_path):
+    target_model.fit_generator(
+        train_datagen.flow(),
+        steps_per_epoch=g_steps_per_epoch,
+        epochs=g_epochs, verbose=1, callbacks=callbacks
+    )
+    target_model.save_weights(save_weight_path)
+
+def test(load_weight_path):
+    print('[reid] benchmark ...')
+    model.load_weights(load_weight_path)
+    e = Evaluator(dataset, feat_model, g_img_h, g_img_w)
+    e.compute()
+
 # #################################################################################
 
 # check list