Spaces:
Runtime error
Runtime error
File size: 5,909 Bytes
48af67d 8bf6199 57f31e9 8bf6199 12880dc 8bf6199 12880dc 8bf6199 12880dc |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 |
---
title: YoloV3 on PASCAL VOC Dataset From Scratch (Slide for GradCam output)
emoji: 🚀
colorFrom: gray
colorTo: blue
sdk: gradio
sdk_version: 3.39.0
app_file: app.py
pinned: false
license: mit
---
Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
# [GithubREPO](https://github.com/deepanshudashora/ERAV1/tree/master/session13)
# Training Procedure
1. The model is trained on Tesla T4 (15GB GPU memory)
2. The training is completed in two phases
3. The first phase contains 20 epochs and second phase contains another 20 epochs
4. In the first training we see loss dropping correctly but in the second training it drops less
5. We run our two training loops separately and do not run any kind of validation on them, except for validation loss
6. Later we evaluate the model and get the numbers
7. The lightning generally saves the model as .ckpt format, so we convert it to torch format by saving state dict as .pt format
8. For doing this we use these two lines of code
```
best_model = torch.load(weights_path)
torch.save(best_model['state_dict'], f'best_model.pth')
litemodel = YOLOv3(num_classes=num_classes)
litemodel.load_state_dict(torch.load("best_model.pth",map_location='cpu'))
device = "cpu"
torch.save(litemodel.state_dict(), PATH)
```
8. The model starts overfitting on the dataset after 30 epochs
9. Future Improvements
1. Train the model in 1 shot instead of two different phases
2. Keep a better batch size (Basically earn more money and buy a good GPU)
3. Data transformation also plays a vital role here
4. OneCycle LR range needs to be appropriately modified for a better LR
# Data Transformation
Along with the transforms mentioned in the [config file](https://github.com/deepanshudashora/ERAV1/blob/master/session13/lightning_version/config.py), we also apply **mosaic transform** on 75% images
[Reference](https://www.kaggle.com/code/nvnnghia/awesome-augmentation/notebook)
# Accuracy Report
```
Class accuracy is: 82.999725%
No obj accuracy is: 96.828300%
Obj accuracy is: 76.898473%
MAP: 0.29939851760864258
```
# [Training Logs](https://github.com/deepanshudashora/ERAV1/blob/master/session13/lightning_version/merged_logs.csv)
#### For faster execution we run the validation step after 20 epochs for the first 20 epochs of training and after that after every 5 epochs till 40 epochs
```
Unnamed: 0 lr-Adam step train_loss epoch val_loss
6576 6576 NaN 164299 4.186745 39.0 NaN
6577 6577 0.000132 164349 NaN NaN NaN
6578 6578 NaN 164349 2.936086 39.0 NaN
6579 6579 0.000132 164399 NaN NaN NaN
6580 6580 NaN 164399 4.777130 39.0 NaN
6581 6581 0.000132 164449 NaN NaN NaN
6582 6582 NaN 164449 3.139145 39.0 NaN
6583 6583 0.000132 164499 NaN NaN NaN
6584 6584 NaN 164499 4.596097 39.0 NaN
6585 6585 0.000132 164549 NaN NaN NaN
6586 6586 NaN 164549 5.587294 39.0 NaN
6587 6587 0.000132 164599 NaN NaN NaN
6588 6588 NaN 164599 4.592830 39.0 NaN
6589 6589 0.000132 164649 NaN NaN NaN
6590 6590 NaN 164649 3.914468 39.0 NaN
6591 6591 0.000132 164699 NaN NaN NaN
6592 6592 NaN 164699 3.180615 39.0 NaN
6593 6593 0.000132 164749 NaN NaN NaN
6594 6594 NaN 164749 5.772174 39.0 NaN
6595 6595 0.000132 164799 NaN NaN NaN
6596 6596 NaN 164799 2.894014 39.0 NaN
6597 6597 0.000132 164849 NaN NaN NaN
6598 6598 NaN 164849 4.473828 39.0 NaN
6599 6599 0.000132 164899 NaN NaN NaN
6600 6600 NaN 164899 6.397766 39.0 NaN
6601 6601 0.000132 164949 NaN NaN NaN
6602 6602 NaN 164949 3.789242 39.0 NaN
6603 6603 0.000132 164999 NaN NaN NaN
6604 6604 NaN 164999 5.182691 39.0 NaN
6605 6605 0.000132 165049 NaN NaN NaN
6606 6606 NaN 165049 4.845749 39.0 NaN
6607 6607 0.000132 165099 NaN NaN NaN
6608 6608 NaN 165099 3.672542 39.0 NaN
6609 6609 0.000132 165149 NaN NaN NaN
6610 6610 NaN 165149 4.230726 39.0 NaN
6611 6611 0.000132 165199 NaN NaN NaN
6612 6612 NaN 165199 4.625024 39.0 NaN
6613 6613 0.000132 165249 NaN NaN NaN
6614 6614 NaN 165249 4.549682 39.0 NaN
6615 6615 0.000132 165299 NaN NaN NaN
6616 6616 NaN 165299 4.040627 39.0 NaN
6617 6617 0.000132 165349 NaN NaN NaN
6618 6618 NaN 165349 4.857126 39.0 NaN
6619 6619 0.000132 165399 NaN NaN NaN
6620 6620 NaN 165399 3.081895 39.0 NaN
6621 6621 0.000132 165449 NaN NaN NaN
6622 6622 NaN 165449 3.945353 39.0 NaN
6623 6623 0.000132 165499 NaN NaN NaN
6624 6624 NaN 165499 3.203420 39.0 NaN
6625 6625 NaN 165519 NaN 39.0 3.081895
```
# Results
## For epochs 0 to 19
## From 19 to 20
## Full training logs for loss
 |