update readme

README.md

@@ -4,7 +4,9 @@ Pytorch implementation of YOLOv3
 <p align="left"><img src="data/innsbruck_result.png" height="160"\>  <img src="data/mountain_result.png" height="160"\></p>
 
 ## What's New
-- **19/05/05 We have verified that our repo exactly reproduces darknet's training using the default configuration, with COCO AP ~= 0.277 on train / val2017.**
+- **19/12/17 Now our repo exactly reproduces the train / eval performance of darknet!**
+- **19/12/17 AP difference of evaluation between darknet and our repo has been eliminated by modifying the postprocess: one-hot class output to multiple-class output.**
+- 19/05/05 We have verified that our repo exactly reproduces darknet's training using the default configuration, with COCO AP ~= 0.277 on train / val2017.
 - 19/02/12 verified inference COCO AP [IoU=0.50:0.95] = 0.297 with val2017, 416x416, batchsize = 8 and w/o random distortion
 - 18/11/27 [COCO AP results of darknet (training) are reproduced with the same training conditions](#performance)
 - 18/11/20 verified inference COCO AP [IoU=0.50:0.95] = 0.302 (paper: 0.310), val5k, 416x416  
@@ -15,8 +17,8 @@ Pytorch implementation of YOLOv3
 #### Inference using yolov3.weights
 <table><tbody>
 <tr><th align="left" bgcolor=#f8f8f8> </th>     <td bgcolor=white> Original (darknet) </td><td bgcolor=white> Ours (pytorch) </td></tr>
-<tr><th align="left" bgcolor=#f8f8f8> COCO AP[IoU=0.50:0.95], inference</th> <td bgcolor=white> 0.310 </td><td bgcolor=white> 0.302 </td></tr>
-<tr><th align="left" bgcolor=#f8f8f8> COCO AP[IoU=0.50],      inference</th> <td bgcolor=white> 0.553 </td><td bgcolor=white> 0.544 </td></tr>
+<tr><th align="left" bgcolor=#f8f8f8> COCO AP[IoU=0.50:0.95], inference</th> <td bgcolor=white> 0.310 </td><td bgcolor=white> 0.311 </td></tr>
+<tr><th align="left" bgcolor=#f8f8f8> COCO AP[IoU=0.50],      inference</th> <td bgcolor=white> 0.553 </td><td bgcolor=white> 0.558 </td></tr>
 </table></tbody>
 
 #### Training
@@ -28,8 +30,8 @@ The models have been evaluated on the COCO 2017 val dataset using our repo.
 <table><tbody><tr><th align="left" bgcolor=#f8f8f8> </th> <td bgcolor=white> darknet weights </td><td bgcolor=white> darknet repo </td><td bgcolor=white> Ours (pytorch) </td><td bgcolor=white> Ours (pytorch) </td></tr>
 <tr><th align="left" bgcolor=#f8f8f8> batchsize </th> <td bgcolor=white> ?? </td><td bgcolor=white> 4 </td><td bgcolor=white> 4 </td> <td bgcolor=white> 8 </td> </tr>
 <tr><th align="left" bgcolor=#f8f8f8> speed [iter/min](*) </th> <td bgcolor=white> ?? </td><td bgcolor=white> <b>19.2</b> </td><td bgcolor=white> <b>19.4</b> </td> <td bgcolor=white> 21.0 </td> </tr>
-<tr><th align="left" bgcolor=#f8f8f8> COCO AP[IoU=0.50:0.95], training</th> <td bgcolor=white> 0.302 </td><td bgcolor=white> <b>0.278</b> </td> <td bgcolor=white> <b>0.277</b> </td> <td bgcolor=white> 0.292 </td> </tr>
-<tr><th align="left" bgcolor=#f8f8f8> COCO AP[IoU=0.50],      training</th> <td bgcolor=white> 0.544 </td><td bgcolor=white> <b>0.476</b> </td> <td bgcolor=white> <b>0.478</b> </td> <td bgcolor=white> 0.500 </td> </tr>
+<tr><th align="left" bgcolor=#f8f8f8> COCO AP[IoU=0.50:0.95], training</th> <td bgcolor=white> 0.311 </td><td bgcolor=white> <b>0.284</b> </td> <td bgcolor=white> <b>0.283</b> </td> <td bgcolor=white> 0.298 </td> </tr>
+<tr><th align="left" bgcolor=#f8f8f8> COCO AP[IoU=0.50],      training</th> <td bgcolor=white> 0.558 </td><td bgcolor=white> <b>0.488</b> </td> <td bgcolor=white> <b>0.491</b> </td> <td bgcolor=white> 0.511 </td> </tr>
 </table></tbody>
 (*) measured on Tesla V100