commit files to HF hub

README.md

@@ -2,7 +2,6 @@
 tags:
 - image-classification
 - pytorch
-- huggingpics
 library_name: generic
 metrics:
 - accuracy

config.json

@@ -6,13 +6,14 @@
   "attention_probs_dropout_prob": 0.0,
   "custom_pipelines": {
     "pair-classification": {
-      "impl": "pair_classification.PairClassificationPipeline",
+      "impl": "pair_classification_pipeline.PairClassificationPipeline",
       "pt": [
         "ViTForImageClassification"
       ],
       "tf": [
         "TFViTForImageClassification"
-      ]
+      ],
+      "type": "image"
     }
   },
   "encoder_stride": 16,
@@ -31,12 +32,12 @@
   "initializer_range": 0.02,
   "intermediate_size": 3072,
   "label2id": {
-    "chk1_fail": 0,
-    "chk1_pass": 1,
-    "chk2_fail": 2,
-    "chk2_pass": 3,
-    "chk3_fail": 4,
-    "chk3_pass": 5
+    "chk1_fail": "0",
+    "chk1_pass": "1",
+    "chk2_fail": "2",
+    "chk2_pass": "3",
+    "chk3_fail": "4",
+    "chk3_pass": "5"
   },
   "layer_norm_eps": 1e-12,
   "model_type": "vit",

pair_classification.py → pair_classification_pipeline.py

@@ -1,8 +1,35 @@
 from torchvision import transforms
+from transformers import ViTFeatureExtractor, ViTForImageClassification
 from transformers import ImageClassificationPipeline
 import torch
 
 
+class PreTrainedPipeline():
+    def __init__(self, path):
+        """
+        Initialize model
+        """
+        # self.processor = feature_extractor = ViTFeatureExtractor.from_pretrained(model_flag)
+        model_flag = 'google/vit-base-patch16-224-in21k'
+        # model_flag = 'google/vit-base-patch16-384'
+        self.feature_extractor = ViTFeatureExtractor.from_pretrained(model_flag)
+        self.model = ViTForImageClassification.from_pretrained(path)
+        self.pipe = PairClassificationPipeline(self.model, feature_extractor=self.feature_extractor)
+        
+    def __call__(self, inputs):
+        """
+        Args:
+            inputs (:obj:`np.array`):
+                The raw waveform of audio received. By default at 16KHz.
+        Return:
+            A :obj:`dict`:. The object return should be liked {"text": "XXX"} containing
+            the detected text from the input audio.
+        """
+        # input_values = self.processor(inputs, return_tensors="pt", sampling_rate=self.sampling_rate).input_values  # Batch size 1
+        # logits = self.model(input_values).logits.cpu().detach().numpy()[0]
+        return self.pipe(inputs)
+
+
 class PairClassificationPipeline(ImageClassificationPipeline):
     pipe_to_tensor = transforms.ToTensor()
     pipe_to_pil = transforms.ToPILImage()

pipeline.py

@@ -1,36 +1,8 @@
 from torchvision import transforms
-from transformers import ViTFeatureExtractor, ViTForImageClassification, ViTConfig
 from transformers import ImageClassificationPipeline
 import torch
 
 
-
-class PreTrainedPipeline():
-    def __init__(self, path):
-        """
-        Initialize model
-        """
-        # self.processor = feature_extractor = ViTFeatureExtractor.from_pretrained(model_flag)
-        model_flag = 'google/vit-base-patch16-224-in21k'
-        # model_flag = 'google/vit-base-patch16-384'
-        self.feature_extractor = ViTFeatureExtractor.from_pretrained(model_flag)
-        self.model = ViTForImageClassification.from_pretrained(path)
-        self.pipe = PairClassificationPipeline(self.model, feature_extractor=self.feature_extractor)
-        
-    def __call__(self, inputs):
-        """
-        Args:
-            inputs (:obj:`np.array`):
-                The raw waveform of audio received. By default at 16KHz.
-        Return:
-            A :obj:`dict`:. The object return should be liked {"text": "XXX"} containing
-            the detected text from the input audio.
-        """
-        # input_values = self.processor(inputs, return_tensors="pt", sampling_rate=self.sampling_rate).input_values  # Batch size 1
-        # logits = self.model(input_values).logits.cpu().detach().numpy()[0]
-        return self.pipe(inputs)
-
-
 class PairClassificationPipeline(ImageClassificationPipeline):
     pipe_to_tensor = transforms.ToTensor()
     pipe_to_pil = transforms.ToPILImage()

pytorch_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b5233e44bcfb172d9200f5f0c63a00ac4411a1783bf6dd42b48dce384a468da1
+oid sha256:a709dd954e0bcc97bd8ed5652becb0709ac5ebc4d36f6e1ae7b15de08019dc01
 size 345635761

runs/events.out.tfevents.1666972137.sa103.11178.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:93ee6a8b55c58e1a4f81c0e7b2484cc7986e796c8cad905ef2417c30f5469136
+size 551