update readme (#1107)

* update readme

* fix test case

* update readme

Author: fsx950223

.github/workflows/python-app.yml

@@ -15,7 +15,7 @@ jobs:
     runs-on: ubuntu-18.04
     strategy:
       matrix:
-        python-version: [3.6, 3.7, 3.8]
+        python-version: [3.7, 3.8, 3.9]
     steps:
     - uses: actions/checkout@v2
     - name: Set up Python ${{ matrix.python-version }}

efficientdet/README.md

@@ -305,6 +305,8 @@ Download data and checkpoints.
 Create a config file for the PASCAL VOC dataset called voc_config.yaml and put this in it.
 
       num_classes: 21
+      lr_warmup_init: 0.08
+      learning_rate: 0.8
       var_freeze_expr: '(efficientnet|fpn_cells|resample_p6)'
       label_map: {1: aeroplane, 2: bicycle, 3: bird, 4: boat, 5: bottle, 6: bus, 7: car, 8: cat, 9: chair, 10: cow, 11: diningtable, 12: dog, 13: horse, 14: motorbike, 15: person, 16: pottedplant, 17: sheep, 18: sofa, 19: train, 20: tvmonitor}
 
@@ -317,7 +319,7 @@ Finetune needs to use --ckpt rather than --backbone_ckpt.
         --model_dir=/tmp/efficientdet-d0-finetune  \
         --ckpt=efficientdet-d0  \
         --train_batch_size=64 \
-        --eval_batch_size=64 --eval_samples=1024 \
+        --eval_batch_size=64 \
         --num_examples_per_epoch=5717 --num_epochs=50  \
         --hparams=voc_config.yaml --val_json_file=tfrecord/json_pascal.json
 
@@ -355,7 +357,7 @@ Finetune needs to use --ckpt rather than --backbone_ckpt.
         --model_dir=/tmp/efficientdet-d0-finetune  \
         --ckpt=efficientdet-d0  \
         --train_batch_size=64 \
-        --eval_batch_size=64 --eval_samples=1024 \
+        --eval_batch_size=64 \
         --num_examples_per_epoch=5717 --num_epochs=50  \
         --hparams=voc_config.yaml
         --strategy=gpus

efficientdet/efficientdet_arch_test.py

@@ -45,72 +45,72 @@ def build_model(self,
     return utils.num_params_flops(False)
 
   def test_efficientdet_d0(self):
-    self.assertSequenceEqual((3880067.0, 2535456183),
+    self.assertSequenceEqual((3880067.0, 2532997127),
                              self.build_model('efficientdet-d0', 512))
 
   def test_efficientdet_d0_channel_first(self):
     self.assertSequenceEqual(
-        (3880067.0, 2534258103),
+        (3880067.0, 2531799047),
         self.build_model('efficientdet-d0', 512, data_format='channels_first'))
 
   def test_efficientdet_d0_511_513(self):
-    self.assertSequenceEqual((3880067.0, 2613160475),
+    self.assertSequenceEqual((3880067.0, 2610687851),
                              self.build_model('efficientdet-d0', (511, 513)))
 
   def test_efficientdet_d1(self):
-    self.assertSequenceEqual((6625898.0, 6101276568),
+    self.assertSequenceEqual((6625898.0, 6095640824),
                              self.build_model('efficientdet-d1', 640))
 
   def test_efficientdet_d1_640_1280(self):
-    self.assertSequenceEqual((6625898.0, 12194901743),
+    self.assertSequenceEqual((6625898.0, 12186897999),
                              self.build_model('efficientdet-d1', (640, 1280)))
 
   def test_efficientdet_d2(self):
-    self.assertSequenceEqual((8097039.0, 10993616292),
+    self.assertSequenceEqual((8097039.0, 10986053156),
                              self.build_model('efficientdet-d2', 768))
 
   def test_efficientdet_d3(self):
-    self.assertSequenceEqual((12032296.0, 24882174639),
+    self.assertSequenceEqual((12032296.0, 24869554579),
                              self.build_model('efficientdet-d3', 896))
 
   def test_efficientdet_d4(self):
-    self.assertSequenceEqual((20723675.0, 55167980877),
+    self.assertSequenceEqual((20723675.0, 55146068329),
                              self.build_model('efficientdet-d4', 1024))
 
   def test_efficientdet_d5(self):
-    self.assertSequenceEqual((33653315.0, 135353202989),
+    self.assertSequenceEqual((33653315.0, 135306619957),
                              self.build_model('efficientdet-d5', 1280))
 
   def test_efficientdet_d6(self):
-    self.assertSequenceEqual((51871782.0, 225532115747),
+    self.assertSequenceEqual((51871782.0, 225467043531),
                              self.build_model('efficientdet-d6', 1280))
 
   def test_efficientdet_d7(self):
-    self.assertSequenceEqual((51871782.0, 324740293607),
+    self.assertSequenceEqual((51871782.0, 324657914255),
                              self.build_model('efficientdet-d7', 1536))
 
   def test_efficientdet_lite0(self):
-    self.assertSequenceEqual((3243470.0, 979428213),
+    self.assertSequenceEqual((3243470.0, 977617221),
                              self.build_model('efficientdet-lite0'))
 
   def test_efficientdet_lite1(self):
-    self.assertSequenceEqual((4248318.0, 1976353506),
+    self.assertSequenceEqual((4248318.0, 1973728714),
                              self.build_model('efficientdet-lite1'))
 
   def test_efficientdet_lite2(self):
-    self.assertSequenceEqual((5252334.0, 3386596870),
+    self.assertSequenceEqual((5252334.0, 3383232590),
                              self.build_model('efficientdet-lite2'))
 
   def test_efficientdet_lite3(self):
-    self.assertSequenceEqual((8350862.0, 7509226979),
+    self.assertSequenceEqual((8350862.0, 7503764515),
                              self.build_model('efficientdet-lite3'))
 
   def test_efficientdet_lite3x(self):
-    self.assertSequenceEqual((9280862.0, 14026487930),
+    self.assertSequenceEqual((9280862.0, 14019809178),
                              self.build_model('efficientdet-lite3x'))
 
   def test_efficientdet_lite4(self):
-    self.assertSequenceEqual((15130894.0, 20232158598),
+    self.assertSequenceEqual((15130894.0, 20221443966),
                              self.build_model('efficientdet-lite4'))
 
 

efficientdet/main.py

@@ -356,7 +356,7 @@ def run_train_and_eval(e):
         if p.exitcode != 0:
           return p.exitcode
       else:
-        tf.compat.v1.reset_default_graph()
+        tf.reset_default_graph()
         run_train_and_eval(e)
 
   else:

efficientdet/tf2/README.md

@@ -218,7 +218,7 @@ You can run inference for a video and show the results online:
     // Run eval.
     !python eval.py  \
         --model_name=${MODEL}  --model_dir=${CKPT_PATH}  \
-        --val_file_pattern=tfrecord/val/pascal*.tfrecord  \
+        --val_file_pattern=tfrecord/val*  \
         --val_json_file=annotations/instances_val2017.json
 
 ## 8. Finetune on PASCAL VOC 2012 with detector COCO ckpt.
@@ -239,6 +239,8 @@ Download data and checkpoints.
 Create a config file for the PASCAL VOC dataset called voc_config.yaml and put this in it.
 
       num_classes: 21
+      lr_warmup_init: 0.08
+      learning_rate: 0.8
       var_freeze_expr: '(efficientnet|fpn_cells|resample_p6)'
       label_map: {1: aeroplane, 2: bicycle, 3: bird, 4: boat, 5: bottle, 6: bus, 7: car, 8: cat, 9: chair, 10: cow, 11: diningtable, 12: dog, 13: horse, 14: motorbike, 15: person, 16: pottedplant, 17: sheep, 18: sofa, 19: train, 20: tvmonitor}