Auto Visualization

auto_run.py

@@ -0,0 +1,148 @@
+#%% 
+import hourglass
+import imp
+imp.reload(hourglass)
+from hourglass import HourglassNet
+from constants import *
+import matplotlib.pyplot as plt
+import os
+
+import data_generator
+imp.reload(data_generator)
+import data_generator
+import time
+
+import re
+import pandas as pd
+import evaluation
+from evaluation import Evaluation
+imp.reload(evaluation)
+from HeatMap import HeatMap
+
+#import HeatMap
+#imp.reload(HeatMap)
+
+import numpy as np
+
+#%% find duplicates in list
+
+def find_dup(pair_list, element):
+    for (a,b) in pair_list:
+        if element == a or element == b:
+            return True
+    return False
+
+#%% 
+
+def find_epochs(base_dir, sub_dir, epoch_dic, visited_models):
+    directory = os.listdir(os.path.join(base_dir, sub_dir))
+    weight_file = [f for f in directory if (f.endswith(".hdf5"))]
+    for name in weight_file:
+        match = re.search('hpe_epoch(\d+)', name)
+        if match:
+            epoch.append((match.group(1), sub_dir))
+
+    resume_file = [f for f in os.listdir(base_dir) if (sub_dir in f and f is not sub_dir)]
+    for file in resume_file:
+        visited_models.append(file)
+        for name in os.listdir(os.path.join(base_dir, file)):
+            match = re.search('hpe_epoch(\d+)', name)
+            #dup = find_dup(epoch, match.group(1))
+            #print("match is ", match, "dup is ", dup)
+            if match and not find_dup(epoch, match.group(1)):
+                epoch.append((match.group(1), file))
+
+    return list(set(epoch))
+
+
+#%% save stacked evaluation heatmaps
+
+def stacked_eval_heatmaps(test_df,eval):
+
+    generator = data_generator.DataGenerator(
+            df=test_df,
+            base_dir=DEFAULT_VAL_IMG_PATH,
+            input_dim=INPUT_DIM,
+            output_dim=OUTPUT_DIM,
+            num_hg_blocks=eval.num_hg_blocks,
+            shuffle=False,  
+            batch_size=len(test_df),
+            online_fetch=False)
+
+    # Select image to predict heatmaps
+    X_batch, y_stacked = generator[0] # There is only one batch in the generator
+    # X_batch, y_stacked = evaluation.load_and_preprocess_img('data/skier.jpg', eval.num_hg_blocks)
+    y_batch = y_stacked[0] # take first hourglass section
+    # Save stacked heatmap images to disk
+    m_batch = test_df.to_dict('records') # TODO: eventually this will be passed from data generator as metadata
+    print("\n\nEval start:   {}\n".format(time.ctime()))
+    eval.visualize_batch(X_batch, y_batch, m_batch)
+    print("\n\nEval end:   {}\n".format(time.ctime()))
+
+#%% 
+
+def predict_kp_bbox(val_df, eval):
+    generator = data_generator.DataGenerator(
+                df=val_df,
+                base_dir=DEFAULT_VAL_IMG_PATH,
+                input_dim=INPUT_DIM,
+                output_dim=OUTPUT_DIM,
+                num_hg_blocks=DEFAULT_NUM_HG,
+                shuffle=False,  
+                batch_size=1,
+                online_fetch=False)
+
+    # Select image to predict heatmaps
+    X_batch, y_stacked = generator[168] # choose one image for evaluation
+    y_batch = y_stacked[0] # take first hourglass section
+    X, y = X_batch[0], y_batch[0] # take first example of batch
+
+    # Get predicted heatmaps for image
+    predict_heatmaps=eval.predict_heatmaps(X_batch)
+
+    # Get predicted keypoints from last hourglass (eval.num_hg_blocks-1)
+    keypoints = eval.heatmaps_to_keypoints(predict_heatmaps[eval.num_hg_blocks-1, 0, :, :, :])
+    print(keypoints)
+    # Get bounding box image from heatmap
+    heatmap = y[:,:,0]
+    hm = HeatMap(X,heatmap)
+    img = np.array(hm.image)
+
+    # Clear plot image
+    plt.clf()
+    # Plot predicted keypoints on bounding box image
+    x = []
+    y = []
+    for i in range(NUM_COCO_KEYPOINTS):
+        if(keypoints[i,0] != 0 and keypoints[i,1] != 0):
+            x.append(keypoints[i,0])
+            y.append(keypoints[i,1])
+    plt.scatter(x,y)
+    plt.imshow(img)
+
+
+# %% find each model's cooresponding epochs 
+
+h = HourglassNet(NUM_COCO_KEYPOINTS,DEFAULT_NUM_HG,INPUT_CHANNELS,INPUT_DIM,OUTPUT_DIM)
+_, val_df = h.load_and_filter_annotations(DEFAULT_TRAIN_ANNOT_PATH,DEFAULT_VAL_ANNOT_PATH,0.1)
+
+representative_set_df = pd.read_pickle(os.path.join(DEFAULT_PICKLE_PATH, 'representative_set.pkl'))
+models = [] # keep track of the models that are visited 
+epoch = [] # extract out all epoch numbers (across different files) from model logs
+output_models = os.listdir(DEFAULT_OUTPUT_BASE_DIR) # keep track of the models have been explored before
+
+for sub_dir in os.listdir(DEFAULT_MODEL_BASE_DIR):
+    if '_hg_' in sub_dir and sub_dir not in models and sub_dir not in output_models:
+        models.append(sub_dir)
+        epoch = find_epochs(DEFAULT_MODEL_BASE_DIR, sub_dir, epoch, models)
+        for (n_epoch, model_file) in epoch:
+            eval = evaluation.Evaluation(
+                    model_sub_dir=model_file,
+                    epoch=int(n_epoch))
+            stacked_eval_heatmaps(representative_set_df,eval)
+            predict_kp_bbox(val_df,eval)
+
+
+
+
+# %%

constants.py

@@ -45,7 +45,7 @@ class ImageAugmentationStrength(Enum):
     DEFAULT_VAL_ANNOT_PATH = 'data/annotations/person_keypoints_val2017.json'
     DEFAULT_TRAIN_IMG_PATH = 'data/coco'
     DEFAULT_VAL_IMG_PATH = 'data/coco'
-    DEFAULT_PICKLE_PATH = './Pickles/'
+    DEFAULT_PICKLE_PATH = 'Pickles'
 
 COCO_KEYPOINT_LABEL_ARR = ["nose","left_eye","right_eye","left_ear","right_ear","left_shoulder","right_shoulder","left_elbow","right_elbow","left_wrist","right_wrist","left_hip","right_hip","left_knee","right_knee","left_ankle","right_ankle"]
 

run_evaluation.py

@@ -7,6 +7,7 @@
 import matplotlib.pyplot as plt
 import os
 
+#%%
 h = HourglassNet(NUM_COCO_KEYPOINTS,DEFAULT_NUM_HG,INPUT_CHANNELS,INPUT_DIM,OUTPUT_DIM)
 _, val_df = h.load_and_filter_annotations(DEFAULT_TRAIN_ANNOT_PATH,DEFAULT_VAL_ANNOT_PATH,0.1)
 
@@ -18,10 +19,10 @@
 imp.reload(HeatMap)
 
 representative_set_df = pd.read_pickle(os.path.join(DEFAULT_PICKLE_PATH, 'representative_set.pkl'))
-subdir = '2021-03-22-20h-23m_batchsize_12_hg_8_loss_weighted_mse_aug_medium_resume_2021-03-25-20h-02m'
+subdir = '2021-03-28-09h-24m_batchsize_16_hg_4_loss_keras_mse_aug_light_sigma4_learningrate_5.0e-03_opt_adam_gt-4kp_activ_linear_subset_0.50_resume_2021-03-28-22h-10m'
 eval = evaluation.Evaluation(
     model_sub_dir=subdir,
-    epoch=43)
+    epoch=30)
 
 # %% Save stacked evaluation heatmaps
 import data_generator