Skip to content

Latest commit

 

History

History
116 lines (93 loc) · 6.21 KB

README.fpn.md

File metadata and controls

116 lines (93 loc) · 6.21 KB

B0-4Y : EfficientNetB0 + 4 YOLO Layers on darknet AlexeyAB

EfficientNetB0 for classifier(AKA Backbone) and 4 YOLO Layers for detector on darknet Framework.

  • enet4y2 with 4.146BFlops to calculate convolution layers
  • 20MB weights file size
  • 20days @ V100 GPU for taining with COCO
  • 47.1 mAP50 with COCO 5K

enet4y2 network structure

The original cfg which on AlexeyAB includes 2 YOLO Layers as detector.
enet4y2 and the original enet-coco have same backbone classifier therefore use same backbone weights.

original README

Object Detector from Dataset download until estimation mAP

Download Dataset

$ mkdir work;cd work
$ ./scripts/get_coco_dataset.sh

Training using backbone weights pre-trained via ImageNet

backbone weigts
This weights of classifier may be generated for Cross Stage Partial Networks.

enet4y2-coco weights
This weights of detector is generated for enet4y2-coco.cfg.

  • write coco.data
classes = 80
train   = work/coco/trainvalno5k.txt
valid   = work/coco/5k.txt
names   = data/coco.names
backup  = backup4y2

  • download backbone weights 'enetb0_final.weights' from Website

  • start training

$ ./darknet detector train coco.data cfg/enet4y2-coco.cfg enetb0_final.weights

Unfortunately wait 20days even if V100 GPGPU.

enet4y2-coco estimation mAP against enet-coco(detector as reference)

Summary via pycocotools

/ AP50-95 AP50 AP75 Small Medium Large Judgement
enet-coco 0.240 0.452 0.231 0.086 0.260 0.369 criterion
enet4y2-coco 0.236 0.471 0.213 0.091 0.258 0.339 target
even win lose win(?) even lose result

enet4y2-coco.cfg pycocotools details

$ ./darknet detector map coco.data cfg/enet4y2-coco.cfg backup4y2/enet4y2-coco_final.weights
$ ./scripts/pycocoeval.py -r results/coco_results.json -g work2/coco/annotations/instances_val2014.json

0.23648480928288304
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.236
 Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.471
 Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.213
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.091
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.258
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.339
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.222
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.346
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.371
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.181
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.401
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.519

progress bar while training
can see training progress bar with ./scripts/avgout.py like below,

loss range: 3002.268311 to 3.872015
-----------------------------------------------------------------------------------------------------------
|  iteration         loss                                       bar                                       |
-----------------------------------------------------------------------------------------------------------
       13983     6.768050 ################################################################################|
       27995     5.574761 ##########################################################################
       42005     5.929069 ##############################################################################
       56015     5.440513 ########################################################################
       70026     5.273204 ######################################################################
       84036     5.494036 #########################################################################
       98046     5.522543 #########################################################################
      112058     5.210263 #####################################################################
      126068     5.341331 #######################################################################
      140081     5.041574 ###################################################################
      154091     5.278211 ######################################################################
      168103     5.333329 #######################################################################
      182114     5.006047 ##################################################################
      196124     4.866174 ################################################################
      210135     4.911802 #################################################################
      224145     4.948373 #################################################################
      238156     4.968527 ##################################################################
      252167     4.585942 #############################################################
      266178     4.541975 ############################################################
      280189     4.465081 ###########################################################
-----------------------------------------------------------------------------------------------------------
MIN   277615     3.872015 >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
-----------------------------------------------------------------------------------------------------------

watch -n 5 'avgout.py log' script is easy way to check status of training everyday.