#!/usr/bin/env python
# coding: utf-8

# # Working with survailance videos, captioning frames and detecting indoor/outdoor

# In[29]:


# This notebook is a tutorial of fastdup beta advanced capabilities. Please sign up for a free version at info@visual-layer.com to unlock those and many other functionalities.


# In[13]:


from IPython.display import Image


# ## Download the fight detection survailance camera dataset

# In[ ]:


get_ipython().system('git clone https://github.com/seymanurakti/fight-detection-surv-dataset.git')


# ## Install fastdup

# In[1]:


get_ipython().run_line_magic('pip', 'install fastdup')
import fastdup


# ## Extract videos into frames

# In[4]:


get_ipython().system('mkdir -p frames')
fastdup.extract_video_frames('fight-detection-surv-dataset/fight/', 'frames/fight')
fastdup.extract_video_frames('fight-detection-surv-dataset/noFight/', 'frames/no-fight')


# ## Quick preview of video frames

# In[8]:


fight_files=get_ipython().getoutput("find frames/fight/ -name '*.jpg'")
no_fight_files=get_ipython().getoutput("find frames/no-fight/ -name '*.jpg'")


# In[9]:


ret = fastdup.generate_sprite_image(fight_files, 55, ".")[0]
from IPython.display import Image
Image(filename=ret)
                              


# In[12]:


ret = fastdup.generate_sprite_image(no_fight_files, 55, ".")[0]
Image(filename=ret)


# ## Build fastdup model

# In[3]:


# Cluster similar frames together
fd = fastdup.create(input_dir='frames', work_dir='out_frames')
fd.run(overwrite=True, model_path='dinov2s')


# In[10]:


fd.vis.component_gallery(num_images=10)


# ## Look at video outliers, and caption them automatically

# In[6]:


fd.vis.outliers_gallery(num_images=10, label_col='blip')


# In[25]:


# In the above reports the captions are given under the "label" field name


# ## Look at image outliers, and classify them into indoors/outdoors

# In[10]:


fd.vis.outliers_gallery(num_images=10, label_col='indoors_outdoors')


# In[28]:


# As can be seen the indoors/outdoors classifier is 90% correct on the above example


# In[ ]: