change gradio interface to use files instead of images. expand valid list to sync with updated model.

app.py

@@ -5,25 +5,41 @@ import numpy as np
 import requests
 import torch
 import heapq
+import argparse
+
 from PIL import Image
 
 from huggingface_hub import snapshot_download
 from huggingface_hub import hf_hub_download
 
-show_all = False
+parser = argparse.ArgumentParser()
+parser.add_argument('--show_all',action = 'store_true')
+parser.add_argument('--local',action = 'store')
+
+args = parser.parse_args()
+show_all = args.show_all
+local = args.local
+
 def readLines(filename):
     with open(filename,'r') as f:
         return(f.read().splitlines())
 
+if local:
+    if len(local) == 0:
+        local = "jaguaridfull"
+    print(f"Using model in {local}")
+    model = tf.saved_model.load(local)
+    labels = readLines('master_labels.txt')
+else:
+    model_path = snapshot_download(repo_id="eshieh2/jaguarid_pantanal")
+    model = tf.saved_model.load(f"{model_path}/saved_model")
+    labels = readLines('labels.txt')
+serving = model.signatures['serving_default']
 
-labels = readLines('labels.txt')
 labels = [x for x in labels if x] #remove blanks
 label_count = len(labels)
 valid =  readLines('valid.txt')
 
-model_path = snapshot_download(repo_id="eshieh2/jaguarid_pantanal")
-model = tf.saved_model.load(f"{model_path}/saved_model")
-serving = model.signatures['serving_default']
 
 detector_path = hf_hub_download(repo_id= "eshieh2/jaguarhead",
                                              filename = "jaguarheadv5.pt")
@@ -31,12 +47,48 @@ detector = torch.hub.load('ultralytics/yolov5', 'custom', path = detector_path)
 
 topk = 3
 
-def classify_image(in_image):
+from PIL import ExifTags
+
+def open_image_correct_orientation(image_path):
+    img = Image.open(image_path)
+
+    # Try to get EXIF orientation data
+    try:
+        for tag in ExifTags.TAGS:
+            if ExifTags.TAGS[tag] == "Orientation":
+                orientation_tag = tag
+                break
+        exif = img._getexif()
+        if exif and orientation_tag in exif:
+            orientation = exif[orientation_tag]
+            new_image = None
+            if orientation == 3:
+                new_image = img.rotate(180, expand=True)
+            elif orientation == 6:
+                new_image = img.rotate(270, expand=True)
+            elif orientation == 8:
+                new_image = img.rotate(90, expand=True)
+            if new_image is not None:
+                # Strip EXIF data by saving without it
+                data = list(new_image.getdata())
+                image_without_exif = Image.new(new_image.mode, new_image.size)
+                image_without_exif.putdata(data)
+                img = image_without_exif
+            
+    except (AttributeError, KeyError, IndexError):
+        pass  # If EXIF data is missing, do nothing
+
+    return img
+  
+def classify_image(in_image,confidence):
+  in_image = open_image_correct_orientation(in_image)
   if in_image is None:
     return None
   width,height = in_image.size
   heads = detector(in_image)
   masks = [] # tuple of box coords and string
+  retLabel = ''
+  index = 1
   for head in heads.xyxy[0]:
     x,y,x2,y2,pct,cls = head.numpy()
     w = x2 - x
@@ -49,11 +101,13 @@ def classify_image(in_image):
     prediction = serving(tf.convert_to_tensor(inp))['output_0']
     prediction = tf.squeeze(prediction)
     pred = {labels[i]: float(prediction[i]) for i in range(label_count)}
+    if confidence is not None:
+      pred = {key: value for key, value in pred.items() if value > confidence}
     #print(pred)
     rect = (int(x),int(y),int(x2),int(y2))
+    label = ''
     if topk is not None:
       top = heapq.nlargest(topk,pred,key=pred.get)
-      label = ''
       for t in top:
         if show_all or t.lower() in valid:
           if len(label) != 0:
@@ -65,14 +119,35 @@ def classify_image(in_image):
     else:
       max_key = max(pred, key=pred.get)
       if show_all or max_key.lower() in valid:
-            masks.append((rect,f"{max_key}:{pred[max_key]}"))
+        label = f"{max_key}:{pred[max_key]}"
       else:
-            masks.append((rect,f"unknown",))
-  return (in_image,masks)
+        label = f"unknown"
+      masks.append((rect,label))
+    if index > 1:
+      retLabel += ", "
+    retLabel += f"[{index} {label}]"
+    index += 1
+      
+  return (in_image,masks),retLabel
+
+def classify_multiple(files,confidence):
+  textResult = ''
+  firstResult = None
+  for file in files:
+    result = classify_image(file,confidence)
+    if firstResult is None:
+      firstResult = result[0]
+    textResult = textResult + f"{os.path.basename(file)}: {result[1]}\n"
+    
+  return firstResult, textResult
+#image = gr.Image(type='pil')
+image = gr.File(label = "Input Images", file_count = 'multiple', file_types = ['image'])
+conf = gr.Radio(choices = [("Best Guess", None), ("Reduce False Positives",0.23)], label = "Confidence")
 
-image = gr.Image(type='pil')
-output = gr.AnnotatedImage()
+output = gr.AnnotatedImage(label = "First Result")
+output2 = gr.Textbox(label = "File Results")
 title = "JaguarID AI identification App"
-desc = "Identifies the following: Medrosa, Guaraci, Marcela, Bagua, Manath, Margo, Ti, Ousado, Saseka, Patricia, Kasimir. Confidence displayed after the name is on a 0.0 - 1.0 scale. Must have a clear front/side head view."
+valid_display = [word.capitalize() for word in valid]
+desc = f"Identifies the following: {', '.join(valid_display)} . Confidence displayed after the name is on a 0.0 - 1.0 scale. Must have a clear front/side head view."
 
-gr.Interface(fn=classify_image, inputs=image, outputs=output, examples = [["medrosa.jpg"],["guaraci.jpg"],["marcela.jpg"]], title = title, description = desc).launch()
+gr.Interface(fn=classify_multiple, inputs=[image,conf], outputs=[output,output2], examples = [[["medrosa.jpg"]],[["guaraci.jpg"]],[["marcela.jpg"]]], title = title, description = desc).launch()

valid.txt

@@ -1,11 +1,18 @@
 bagua
+bernard
+bororo
 guaraci
+ibaca
+jaju
 kasimir
+krishna
+kyyavera
 manath
-margo
 marcela
+margo
 medrosa
 ousado
 patricia
-saseka
+rio
 ti
+xingu