Upgrade to Python 3.8

.gitignore

@@ -10,7 +10,6 @@ db.sqlite3
 __pycache__
 .DS_Store
 tags
-api/migrations/
 api/places365/model/
 Conv2d.patch
 Linear.patch
@@ -31,7 +30,6 @@ api/places365/*.tar.gz
 *.db
 media*
 *.log
-api/migrations_dev
 protected_media
 .vscode
 .coverage

Dockerfile

@@ -1,4 +1,4 @@
-FROM ubuntu:16.04
+FROM ubuntu:20.10
 MAINTAINER Hooram Nam <[email protected]>
 
 ENV MAPZEN_API_KEY mapzen-XXXX
@@ -13,14 +13,14 @@ RUN apt-get update && \
     libxrender-dev \
     wget \
     curl \
-    nginx 
+    nginx \
+    cmake
 
 RUN apt-get install -y bzip2
 
-
 RUN wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
 RUN bash Miniconda3-latest-Linux-x86_64.sh -b -p /miniconda
-# RUN apt-get install libopenblas-dev liblapack-dev
+
 RUN /miniconda/bin/conda install -y faiss-cpu -c pytorch
 RUN /miniconda/bin/conda install -y cython
 
@@ -33,19 +33,24 @@ RUN apt-get update && \
     cmake .. -DDLIB_USE_CUDA=0 -DUSE_AVX_INSTRUCTIONS=0 && \
     cmake --build . && \
     cd /dlib && \
-    /miniconda/bin/python setup.py install --no USE_AVX_INSTRUCTIONS --no DLIB_USE_CUDA 
+    /miniconda/bin/python setup.py install --no USE_AVX_INSTRUCT
+
 
-RUN /miniconda/bin/conda install -y pytorch=0.4.1 -c pytorch
-# RUN /venv/bin/pip install http://download.pytorch.org/whl/cpu/torch-0.4.1-cp35-cp35m-linux_x86_64.whl && /venv/bin/pip install torchvision
+RUN /miniconda/bin/conda install -y pytorch torchvision torchaudio cpuonly -c pytorch
 RUN /miniconda/bin/conda install -y psycopg2
 
+RUN /miniconda/bin/conda install -y numpy -c pytorch
+RUN /miniconda/bin/conda install -y pandas -c pytorch
+RUN /miniconda/bin/conda install -y scikit-learn -c pytorch
+RUN /miniconda/bin/conda install -y scikit-image -c pytorch
+
 RUN mkdir /code
 WORKDIR /code
 COPY requirements.txt /code/
 RUN /miniconda/bin/pip install -r requirements.txt
 
 RUN /miniconda/bin/python -m spacy download en_core_web_sm
-
+RUN apt-get install -y libgl1-mesa-glx
 WORKDIR /code/api/places365
 RUN wget https://s3.eu-central-1.amazonaws.com/ownphotos-deploy/places365_model.tar.gz
 RUN tar xf places365_model.tar.gz
@@ -93,7 +98,6 @@ ENV TIME_ZONE UTC
 EXPOSE 80
 COPY . /code
 
-
 RUN mv /code/config_docker.py /code/config.py
 
 WORKDIR /code

api/autoalbum.py

@@ -9,7 +9,7 @@
 import os
 import shutil
 import numpy as np
-
+import uuid
 import ipdb
 
 from django_rq import job
@@ -20,9 +20,7 @@
 import pytz
 
 @job
-def regenerate_event_titles(user):
-    job_id = rq.get_current_job().id
-
+def regenerate_event_titles(user,job_id):
     if LongRunningJob.objects.filter(job_id=job_id).exists():
         lrj = LongRunningJob.objects.get(job_id=job_id)
         lrj.started_at = datetime.now().replace(tzinfo=pytz.utc)
@@ -35,11 +33,7 @@ def regenerate_event_titles(user):
             started_at=datetime.now().replace(tzinfo=pytz.utc),
             job_type=LongRunningJob.JOB_GENERATE_AUTO_ALBUM_TITLES)
         lrj.save()
-
-
-
     try:
-
         aus = AlbumAuto.objects.filter(owner=user).prefetch_related('photos')
         target_count = len(aus)
         for idx,au in enumerate(aus):
@@ -74,9 +68,7 @@ def regenerate_event_titles(user):
 
 
 @job
-def generate_event_albums(user):
-    job_id = rq.get_current_job().id
-
+def generate_event_albums(user, job_id):
     if LongRunningJob.objects.filter(job_id=job_id).exists():
         lrj = LongRunningJob.objects.get(job_id=job_id)
         lrj.started_at = datetime.now().replace(tzinfo=pytz.utc)

api/directory_watcher.py

@@ -2,44 +2,18 @@
 import datetime
 import hashlib
 import pytz
-import time
-import traceback
-from joblib import Parallel, delayed
-import multiprocessing
-
-from api.models import (Photo, Person, LongRunningJob)
-
-from tqdm import tqdm
-from config import image_dirs
+from api.models import (Photo, LongRunningJob)
 
 import api.util as util
 from api.image_similarity import build_image_similarity_index
 
-import ipdb
 from django_rq import job
-import time
-import numpy as np
-import rq
-
-
 from django.db.models import Q
 import json
 
-
-def is_new_image(existing_hashes, image_path):
-    hash_md5 = hashlib.md5()
-    with open(image_path, "rb") as f:
-        for chunk in iter(lambda: f.read(4096), b""):
-            hash_md5.update(chunk)
-    image_hash = hash_md5.hexdigest()
-    if image_hash not in existing_hashes or (
-            not Photo.objects.filter(image_path=image_path).exists()):
-        return image_path
-    return
-
-
 def handle_new_image(user, image_path, job_id):
-    if image_path.lower().endswith('.jpg'):
+
+    if image_path.lower().endswith('.jpg') or image_path.lower().endswith('.jpeg') :
         try:
             elapsed_times = {
                 'md5':None,
@@ -81,69 +55,53 @@ def handle_new_image(user, image_path, job_id):
                     added_on=datetime.datetime.now().replace(tzinfo=pytz.utc),
                     geolocation_json={})
                 #photo._generate_md5()
-
+
+
                 start = datetime.datetime.now()
                 photo._generate_thumbnail()
                 elapsed = (datetime.datetime.now() - start).total_seconds()
                 elapsed_times['thumbnails'] = elapsed
-#                 util.logger.info('thumbnail get took %.2f' % elapsed)
+                util.logger.info('thumbnail get took %.2f' % elapsed)
 
                 start = datetime.datetime.now()
                 photo._generate_captions()
                 elapsed = (datetime.datetime.now() - start).total_seconds()
                 elapsed_times['captions'] = elapsed
-#                 util.logger.info('caption generation took %.2f' % elapsed)
+                util.logger.info('caption generation took %.2f' % elapsed)
 
 #                 start = datetime.datetime.now()
 #                 photo._save_image_to_db()
 #                 elapsed = (datetime.datetime.now() - start).total_seconds()
 #                 elapsed_times['image_save'] = elapsed
-#                 util.logger.info('image save took %.2f' % elapsed)
+                util.logger.info('image save took %.2f' % elapsed)
 
                 start = datetime.datetime.now()
                 photo._extract_exif()
-                photo.save()
-                elapsed = (datetime.datetime.now() - start).total_seconds()
-                elapsed_times['exif'] = elapsed
-#                 util.logger.info('exif extraction took %.2f' % elapsed)
-
-                start = datetime.datetime.now()
+                util.logger.info('add to AlbumPlace took %.2f' % elapsed)
                 photo._geolocate_mapbox()
-                photo.save()
-                elapsed = (datetime.datetime.now() - start).total_seconds()
-                elapsed_times['geolocation'] = elapsed
-#                 util.logger.info('geolocation took %.2f' % elapsed)
-
-                start = datetime.datetime.now()
-                photo._add_to_album_place()
-                photo.save()
-                elapsed = (datetime.datetime.now() - start).total_seconds()
-                elapsed_times['album_place'] = elapsed
-#                 util.logger.info('add to AlbumPlace took %.2f' % elapsed)
-
                 start = datetime.datetime.now()
                 photo._extract_faces()
                 elapsed = (datetime.datetime.now() - start).total_seconds()
                 elapsed_times['faces'] = elapsed
-#                 util.logger.info('face extraction took %.2f' % elapsed)
+                util.logger.info('face extraction took %.2f' % elapsed)
 
                 start = datetime.datetime.now()
                 photo._add_to_album_date()
                 elapsed = (datetime.datetime.now() - start).total_seconds()
                 elapsed_times['album_date'] = elapsed
-#                 util.logger.info('adding to AlbumDate took %.2f' % elapsed)
+                util.logger.info('adding to AlbumDate took %.2f' % elapsed)
 
                 start = datetime.datetime.now()
                 photo._add_to_album_thing()
                 elapsed = (datetime.datetime.now() - start).total_seconds()
                 elapsed_times['album_thing'] = elapsed
-#                 util.logger.info('adding to AlbumThing took %.2f' % elapsed)
+                util.logger.info('adding to AlbumThing took %.2f' % elapsed)
 
                 start = datetime.datetime.now()
                 photo._im2vec()
                 elapsed = (datetime.datetime.now() - start).total_seconds()
                 elapsed_times['im2vec'] = elapsed
-#                 util.logger.info('im2vec took %.2f' % elapsed)
+                util.logger.info('im2vec took %.2f' % elapsed)
 
                 util.logger.info("job {}: image processed: {}, elapsed: {}".format(job_id,img_abs_path,json.dumps(elapsed_times)))
 
@@ -154,17 +112,15 @@ def handle_new_image(user, image_path, job_id):
 
         except Exception as e:
             try:
-                util.logger.error("job {}: could not load image {}. reason: {}".format(
+                util.logger.exception("job {}: could not load image {}. reason: {}".format(
                     job_id,image_path, str(e)))
             except:
-                util.logger.error("job {}: could not load image {}".format(job_id,image_path))
-    return
-
+                util.logger.exception("job {}: could not load image {}".format(job_id,image_path))
+
 
+#job is currently not used, because the model.eval() doesn't execute when it is running as a job
 @job
-def scan_photos(user):
-    job_id = rq.get_current_job().id
-
+def scan_photos(user, job_id):
     if LongRunningJob.objects.filter(job_id=job_id).exists():
         lrj = LongRunningJob.objects.get(job_id=job_id)
         lrj.started_at = datetime.datetime.now().replace(tzinfo=pytz.utc)
@@ -178,9 +134,6 @@ def scan_photos(user):
             job_type=LongRunningJob.JOB_SCAN_PHOTOS)
         lrj.save()
 
-
-
-
     added_photo_count = 0
     already_existing_photo = 0
 
@@ -194,7 +147,7 @@ def scan_photos(user):
 
         image_paths = [
             p for p in image_paths
-            if p.lower().endswith('.jpg') and 'thumb' not in p.lower()
+            if (p.lower().endswith('.jpg') or p.lower().endswith('.jpeg')) and 'thumb' not in p.lower()
         ]
         image_paths.sort()
 
@@ -216,26 +169,21 @@ def scan_photos(user):
                 }
             }
             lrj.save()
-        '''
-        image_paths_to_add = Parallel(n_jobs=multiprocessing.cpu_count(), backend="multiprocessing")(delayed(is_new_image)(existing_hashes, image_path) for image_path in tqdm(image_paths)) 
-        image_paths_to_add = filter(None, image_paths_to_add)
-        Parallel(n_jobs=multiprocessing.cpu_count(), backend="multiprocessing")(delayed(handle_new_image)(user, image_path) for image_path in tqdm(image_paths_to_add)) 
-        '''
 
         util.logger.info("Added {} photos".format(len(image_paths_to_add)))
         build_image_similarity_index(user)
 
-        lrj = LongRunningJob.objects.get(job_id=rq.get_current_job().id)
+        lrj = LongRunningJob.objects.get(job_id=job_id)
         lrj.finished = True
         lrj.finished_at = datetime.datetime.now().replace(tzinfo=pytz.utc)
         prev_result = lrj.result
         next_result = prev_result
         next_result['new_photo_count'] = added_photo_count
         lrj.result = next_result
         lrj.save()
-    except Exception as e:
-        util.logger.error(str(e))
-        lrj = LongRunningJob.objects.get(job_id=rq.get_current_job().id)
+    except Exception:
+        util.logger.exception("An error occured:")
+        lrj = LongRunningJob.objects.get(job_id=job_id)
         lrj.finished = True
         lrj.failed = True
         lrj.finished_at = datetime.datetime.now().replace(tzinfo=pytz.utc)

api/face_classify.py

@@ -3,26 +3,12 @@
 from api.models import LongRunningJob
 from api.util import logger
 
-import base64
-import pickle
-import itertools
-import ipdb
-
-from scipy import linalg
 from sklearn.decomposition import PCA
 import numpy as np
-from sklearn import cluster
-from sklearn import mixture
-from scipy.spatial import distance
-from sklearn.preprocessing import StandardScaler
-from sklearn.linear_model import SGDClassifier
 from sklearn.neural_network import MLPClassifier
-from sklearn import svm
-from sklearn.manifold import TSNE
 
 import seaborn as sns
 from django_rq import job
-import rq
 import pytz
 
 import datetime
@@ -69,9 +55,7 @@ def cluster_faces(user):
 
 
 @job
-def train_faces(user):
-    job_id = rq.get_current_job().id
-
+def train_faces(user, job_id):
     if LongRunningJob.objects.filter(job_id=job_id).exists():
         lrj = LongRunningJob.objects.get(job_id=job_id)
         lrj.started_at = datetime.datetime.now().replace(tzinfo=pytz.utc)
@@ -125,25 +109,13 @@ def train_faces(user):
             [f['encoding'] for f in id2face_known.values()])
         person_names_known = np.array(
             [f['person_name'] for f in id2face_known.values()])
-
-        n_clusters = len(set(person_names_known.tolist()))
-
-        # clf = SGDClassifier(loss='log',penalty='l2')
+        logger.info("Before fitting")        
         clf = MLPClassifier(
-            solver='adam', alpha=1e-5, random_state=1, max_iter=1000)
-        # clf = svm.SVC(kernel='linear')
-        # scaler = StandardScaler()
-        # scaler.fit(face_encodings_all)
-        # X = scaler.transform(face_encodings_known)
-        X = face_encodings_known
-        Y = person_names_known
-        clf.fit(X, person_names_known)
-
+            solver='adam', alpha=1e-5, random_state=1, max_iter=1000).fit(face_encodings_known, person_names_known)
+        logger.info("After fitting")    
         face_encodings_unknown = np.array(
             [f['encoding'] for f in id2face_unknown.values()])
-        face_paths_unknown = [
-            f['image_path'] for f in id2face_unknown.values()
-        ]
+
         face_ids_unknown = [f['id'] for f in id2face_unknown.values()]
         pred = clf.predict(face_encodings_unknown)
         probs = np.max(clf.predict_proba(face_encodings_unknown), 1)
@@ -166,9 +138,6 @@ def train_faces(user):
             }
             lrj.save()
 
-#         res = cluster_faces()
-#         print(res)
-
         lrj.finished = True
         lrj.failed = False
         lrj.finished_at = datetime.datetime.now()