peasant98/dronet-pytorch

Dronet, adapted for Pytorch.

Matt Strong

Hi!

I'm Matt (司马修), and I'm a CS PhD student (Robotics+AI) at Stanford University. I work on dexterous manipulation.

• 🤖 During my undergrad at the University of Colorado Boulder, I worked in the HIRO group as an undergraduate researcher, advised by Professor Alessandro Roncone. I was also a researcher in the SBS lab, advised by Professor Wangda Zuo. Along the way, I interned a couple of times at Microsoft. After graduating, I worked at Microsoft for ~2 years on the Customer Experience Platform.
• 🌱 My research involves enabling perception and autonomy for physical human-robot interaction. Previously, on the side, I worked on leveraging machine learning as applied to building energy modeling.
• 🏃 I enjoy running, trail running, biking, swimming, and hiking ⛰️. Check out my Strava!
• ⚡ Languages: Python, C++, Typescript, C#, 中文 (in progress)

dronet-pytorch's Issues

Problem on the converge of collision value

Hello, thanks for your great work on the PyTorch implementation of Dronet :D
In the training progress, I actually come with some problems and wonder if it is my fault.
I downloaded the Udacity steering dataset and the collision dataset collected on bikes, by running the move_data.py and do some small tweaks to the structure of folders, I made the dronet_torch_train work. However, the training result is not ideal. The loss plot shows that the steering value converged at the end, while the training result of the collision data looks not so good.

I have been trying to resolve the issue for days, but I still not find what I did wrong. Could you please offer me some advice on it?
Thanks:)

The structure of the "data" folder is as follows:

Setup Data Collection Properly

Reminder to myself to setup (balanced) data collection for training the model!

Problem on the converge of collision value

Hello, thanks for your great work on the PyTorch implementation of Dronet :D
In the training progress, I actually come with some problems and wonder if it is my fault.
I downloaded the Udacity steering dataset and the collision dataset collected on bikes, by running the move_data.py and do some small tweaks to the structure of folders, I made the dronet_torch_train work. However, the training result is not ideal. The loss plot shows that the steering value converged at the end, while the training result of the collision data looks not so good.

I have been trying to resolve the issue for days, but I still not find what I did wrong. Could you please offer me some advice on it?
Thanks:)

The structure of the "data" folder is as follows:
. -- collision
-- collision_dataset |-- testing | |-- DSCN2571 | | -- images
| |-- GOPR0200
| | -- images | |-- GOPR0255 | | -- images
| |-- GOPR0265
| | -- images | |-- GOPR0366 | | -- images
| |-- GOPR0369
| | -- images | |-- GOPR0382 | | -- images
| |-- GOPR0386
| | -- images | -- UDACITY_TESTING
| -- images |-- training | |-- DSCN2561 | | -- images
| |-- DSCN2562
| | -- images | |-- DSCN2563 | | -- images
| |-- DSCN2564
| | -- images | |-- DSCN2565 | | -- images
| |-- DSCN2566
| | -- images | |-- DSCN2567 | | -- images
| |-- DSCN2568
| | -- images | |-- DSCN2569 | | -- images
| |-- DSCN2573
| | -- images | |-- DSCN2574 | | -- images
| |-- DSCN2575
| | -- images | |-- DSCN2677 | | -- images
| |-- DSCN2678
| | -- images | |-- DSCN2679 | | -- images
| |-- DSCN2680
| | -- images | |-- DSCN2681 | | -- images
| |-- DSCN2683
| | -- images | |-- DSCN2684 | | -- images
| |-- DSCN2685
| | -- images | |-- DSCN2688 | | -- images
| |-- DSCN2689
| | -- images | |-- DSCN2692 | | -- images
| |-- DSCN2693
| | -- images | |-- DSCN2694 | | -- images
| |-- DSCN2697
| | -- images | |-- GOPR0201 | | -- images
| |-- GOPR0203
| | -- images | |-- GOPR0204 | | -- images
| |-- GOPR0212
| | -- images | |-- GOPR0213 | | -- images
| |-- GOPR0214
| | -- images | |-- GOPR0216 | | -- images
| |-- GOPR0217
| | -- images | |-- GOPR0219 | | -- images
| |-- GOPR0220
| | -- images | |-- GOPR0221 | | -- images
| |-- GOPR0223
| | -- images | |-- GOPR0224 | | -- images
| |-- GOPR0225
| | -- images | |-- GOPR0226 | | -- images
| |-- GOPR0228
| | -- images | |-- GOPR0229 | | -- images
| |-- GOPR0230
| | -- images | |-- GOPR0231 | | -- images
| |-- GOPR0232
| | -- images | |-- GOPR0233 | | -- images
| |-- GOPR0234
| | -- images | |-- GOPR0235 | | -- images
| |-- GOPR0236
| | -- images | |-- GOPR0237 | | -- images
| |-- GOPR0238
| | -- images | |-- GOPR0239 | | -- images
| |-- GOPR0240
| | -- images | |-- GOPR0241 | | -- images
| |-- GOPR0242
| | -- images | |-- GOPR0243 | | -- images
| |-- GOPR0244
| | -- images | |-- GOPR0245 | | -- images
| |-- GOPR0246
| | -- images | |-- GOPR0247 | | -- images
| |-- GOPR0249
| | -- images | |-- GOPR0250 | | -- images
| |-- GOPR0251
| | -- images | |-- GOPR0252 | | -- images
| |-- GOPR0253
| | -- images | |-- GOPR0254 | | -- images
| |-- GOPR0256
| | -- images | |-- GOPR0259 | | -- images
| |-- GOPR0260
| | -- images | |-- GOPR0261 | | -- images
| |-- GOPR0262
| | -- images | |-- GOPR0263 | | -- images
| |-- GOPR0264
| | -- images | |-- GOPR0266 | | -- images
| |-- GOPR0267
| | -- images | |-- GOPR0268 | | -- images
| |-- GOPR0269
| | -- images | |-- GOPR0270 | | -- images
| |-- GOPR0271
| | -- images | |-- GOPR0272 | | -- images
| |-- GOPR0273
| | -- images | |-- GOPR0274 | | -- images
| |-- GOPR0275
| | -- images | |-- GOPR0276 | | -- images
| |-- GOPR0277
| | -- images | |-- GOPR0278 | | -- images
| |-- GOPR0297
| | -- images | |-- GOPR0298 | | -- images
| |-- GOPR0299
| | -- images | |-- GOPR0300 | | -- images
| |-- GOPR0301
| | -- images | |-- GOPR0302 | | -- images
| |-- GOPR0303
| | -- images | |-- GOPR0347 | | -- images
| |-- GOPR0349
| | -- images | |-- GOPR0350 | | -- images
| |-- GOPR0351
| | -- images | |-- GOPR0352 | | -- images
| |-- GOPR0353
| | -- images | |-- GOPR0354 | | -- images
| |-- GOPR0356
| | -- images | |-- GOPR0357 | | -- images
| |-- GOPR0358
| | -- images | |-- GOPR0359 | | -- images
| |-- GOPR0360
| | -- images | |-- GOPR0361 | | -- images
| |-- GOPR0362
| | -- images | |-- GOPR0363 | | -- images
| |-- GOPR0364
| | -- images | |-- GOPR0365 | | -- images
| |-- GOPR0367
| | -- images | |-- GOPR0368 | | -- images
| |-- GOPR0370
| | -- images | |-- GOPR0371 | | -- images
| |-- GOPR0373
| | -- images | |-- GOPR0374 | | -- images
| |-- GOPR0375
| | -- images | |-- GOPR0376 | | -- images
| |-- GOPR0377
| | -- images | |-- GOPR0378 | | -- images
| |-- GOPR0379
| | -- images | |-- GOPR0381 | | -- images
| |-- GOPR0383
| | -- images | |-- GOPR0384 | | -- images
| |-- GOPR0385
| | -- images | |-- GOPR0387 | | -- images
| -- UDACITY_TRAINING | -- images
-- validation |-- DSCN2682 | -- images
|-- GOPR0227
| -- images -- UDACITY_VALIDATION
`-- images

285 directories
`