feat: graffiti.org conditioning

README.md

@@ -3,14 +3,16 @@ dataset_info:
   features:
   - name: image
     dtype: image
+  - name: conditioning_image
+    dtype: image
   - name: text
     dtype: string
   splits:
   - name: train
-    num_bytes: 690468770
-    num_examples: 77487
-  download_size: 11106922968
-  dataset_size: 690468770
+    num_bytes: 10454347970
+    num_examples: 77478
+  download_size: 16111025975
+  dataset_size: 10454347970
 ---
 # Dataset Card for Graffiti
 

cli.py

@@ -11,8 +11,7 @@ import os
 from controlnet_aux.processor import Processor
 
 from PIL import UnidentifiedImageError
-from PIL import Image, ImageFile
-ImageFile.LOAD_TRUNCATED_IMAGES = True
+from PIL import Image
 
 
 class Scraper(object):
@@ -194,8 +193,8 @@ class CLI():
             path = Path("./images").rglob("*.jpg")
             for i, img_p in enumerate(path):
                 try:
-                    Image.open(img_p)
-                except UnidentifiedImageError:
+                    Image.open(img_p).load()
+                except Exception:
                     path_name = str(img_p)
                     print(path_name + " is broken. Deleting!")
                     os.remove(path_name)
@@ -233,7 +232,7 @@ class CLI():
                     data = json.loads(row)
                     with open('./images/' + data["file"] + '.json', 'w') as j:
                         j.write(json.dumps(data, indent=2, ensure_ascii=False))
-        elif args.command == 'condition':
+        elif args.command == 'conditioning':
             processor_id = 'softedge_hed'
             processor = Processor(processor_id)
 

data/graffiti.org/conditioning.tar.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:08b04aa1897e3a4c1e25b732cfb4356f91a8103ff6f3487d63acc4d991d6db8a
+size 2587653112

data/graffiti.org/images.tar.gz

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:97cb1cd4fc428561c70dd620036fea594e3d38aeef2037b40bd253fa43aebd19
-size 3213661340
+oid sha256:3f7d02c10631c1c8964f6c4f824ad6df152abe46602073e2884341b37ecf8d23
+size 3211879389

data/graffiti.org/metadata.jsonl

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b7b1c382ba1ea07f6107d8b3a813a15f96fe87a74773dea75ce74796314e8449
-size 6372995
+oid sha256:353d4454f0d12c698a5509a8f3064e0a5ad2f934026853b900156eda3fe8ee7e
+size 6370190

graffiti.py

@@ -71,6 +71,7 @@ class Graffiti(datasets.GeneratorBasedBuilder):
             features=datasets.Features(
                 {
                     "image": datasets.Image(),
+                    "conditioning_image": datasets.Image(),
                     "text": datasets.Value("string")
                 }
             ),
@@ -85,8 +86,10 @@ class Graffiti(datasets.GeneratorBasedBuilder):
         """Returns SplitGenerators."""
         images = []
         metadata = []
+        conditioning = []
         for source in _SOURCES:
             images.append(dl_manager.iter_archive(dl_manager.download("./data/{0}/images.tar.gz".format(source))))
+            conditioning.append(dl_manager.iter_archive(dl_manager.download("./data/{0}/conditioning.tar.gz".format(source))))
             metadata.append(dl_manager.download("./data/{0}/metadata.jsonl".format(source)))
         return [
             datasets.SplitGenerator(
@@ -95,27 +98,35 @@ class Graffiti(datasets.GeneratorBasedBuilder):
                 gen_kwargs={
                     "images": images,
                     "metadata": metadata,
+                    "conditioning": conditioning
                 },
             )
         ]
 
-    def _generate_examples(self, metadata, images):
+    def _generate_examples(self, metadata, images, conditioning):
         idx = 0
         for index, meta in enumerate(metadata):
             m = []
             with open(meta, encoding="utf-8") as f:
                 for row in f:
                     m.append(json.loads(row))
+            c = iter(conditioning[index])
             for file_path, file_obj in images[index]:
                 data = [x for x in m if file_path.endswith(x["file"])][0]
+
+                conditioning_file = next(c)
+                conditioning_file_path = conditioning_file[0]
+                conditioning_file_obj = conditioning_file[1]
+
                 text = data["caption"]
                 if data["artist"] != None:
-                    text += ", with text \"" + data["artist"] + "\""
-                    text += ", in the art of \"" + data["artist"] + "\""
+                    # text += ", with text " + data["artist"]
+                    text += ", in the art of " + data["artist"]
                 if data["city"] != None:
                     text += ", located in " + data["city"]
                 yield idx, {
                         "image": {"path": file_path, "bytes": file_obj.read()},
-                        "text": text
+                        "conditioning_image": {"path": conditioning_file_path, "bytes": conditioning_file_obj.read()},
+                        "text": text,
                     }
                 idx+=1

prepare.sh

@@ -28,5 +28,5 @@ test () {
   datasets-cli test graffiti.py --save_info --all_config
 }
 
-# list ${args[1]}
+list ${args[1]}
 tar ${args[0]} ${args[1]}

requirements.txt

@@ -1,4 +1,6 @@
 bs4
 requests
 transformers
-controlnet-aux==0.0.6
+controlnet-aux==0.0.6
+mediapipe
+datasets