Testing accuracy is too low on ResNET 18 pre tarined model

waseemsofficial · July 19, 2022, 1:11pm

I am using Dusty-nv Jetson-inference. I followed Walhtrough to train the classification of images.
I am training tomato ripening stages. I trained 2 classes the most extreme pure green and pure red. On testing, Results showed only one class and classify all images as pure green ( Pure green 15 images as pure green and pure red 15 test images as pure green again). How to train my data set. I trained up to 35 Epochs.

The other case is that I have 6 classes from pure green to pure red and I trained them to 100 and then 200 epochs. but accuracy is only 20%.

Please identify the issue, and what could be wrong?

AastaLLL · July 20, 2022, 4:35am

Hi,

Did you follow the tutorial below?
If yes, could you share the training log that contains loss and accuracy with us?

github.com

dusty-nv/jetson-inference/blob/master/docs/pytorch-cat-dog.md

<img src="https://github.com/dusty-nv/jetson-inference/raw/master/docs/images/deep-vision-header.jpg" width="100%">
<p align="right"><sup><a href="pytorch-transfer-learning.md">Back</a> | <a href="pytorch-plants.md">Next</a> | </sup><a href="../README.md#hello-ai-world"><sup>Contents</sup></a>
<br/>
<sup>Transfer Learning - Classification</sup></s></p>

# Re-training on the Cat/Dog Dataset

The first model that we'll be re-training is a simple model that recognizes two classes:  cat or dog.

<img src="https://github.com/dusty-nv/jetson-inference/raw/python/docs/images/pytorch-cat-dog.jpg" width="700">

Provided below is an 800MB dataset that includes 5000 training images, 1000 validation images, and 200 test images, each evenly split between the cat and dog classes.  The set of training images is used for transfer learning, while the validation set is used to evaluate classification accuracy during training, and the test images are to be used by us after training completes.  The network is never directly trained on the validation and test sets, only the training set.

The images from the dataset are made up of many different breeds of dogs and cats, including large felines like tigers and mountain lions since the amount of cat images available was a bit lower than dogs.  Some of the images also picture humans, which the detector is essentially trained to ignore as background and focus on the cat vs. dog content.

To get started, first make sure that you have [PyTorch installed](pytorch-transfer-learning.md#installing-pytorch) on your Jetson, then download the dataset below and kick off the training script.  After that, we'll test the re-trained model in TensorRT on some static images and a live camera feed. 

## Downloading the Data

During this tutorial, we'll store the datasets on the host device under `jetson-inference/python/training/classification/data`, which is one of the directories that is automatically [mounted into the container](aux-docker.md#mounted-data-volumes).  This way the dataset won't be lost when you shutdown the container.

This file has been truncated. show original

Thanks.

waseemsofficial · July 20, 2022, 11:30am

Dear, Many thanks for your reply.
Yes, I have followed the exact tutorial as I am doing transfer learning.

I am sharing data for 6 classes, from pure green to Red.
Following is epoch no 21st.

Epoch: [20][  0/900]  Time 151.736 (151.736)  Data 14.413 (14.413)  Loss 1.9510e+00 (1.9510e+00)  Acc@1   0.00 (  0.00)  Acc@5 100.00 (100.00)
Epoch: [20][ 10/900]  Time  0.159 (14.526)  Data  0.000 ( 1.766)  Loss 2.0621e+00 (2.0905e+00)  Acc@1   0.00 (  0.00)  Acc@5 100.00 ( 90.91)
Epoch: [20][ 20/900]  Time  0.157 ( 7.791)  Data  0.000 ( 1.039)  Loss 1.0253e+00 (1.9111e+00)  Acc@1 100.00 ( 19.05)  Acc@5 100.00 ( 95.24)
Epoch: [20][ 30/900]  Time  0.158 ( 5.330)  Data  0.000 ( 0.708)  Loss 1.5839e+00 (1.9153e+00)  Acc@1   0.00 ( 16.13)  Acc@5 100.00 ( 93.55)
Epoch: [20][ 40/900]  Time  0.159 ( 4.069)  Data  0.000 ( 0.537)  Loss 2.4524e+00 (2.0084e+00)  Acc@1   0.00 ( 17.07)  Acc@5 100.00 ( 87.80)
Epoch: [20][ 50/900]  Time  0.158 ( 3.302)  Data  0.001 ( 0.434)  Loss 2.1248e+00 (2.0258e+00)  Acc@1   0.00 ( 17.65)  Acc@5 100.00 ( 90.20)
Epoch: [20][ 60/900]  Time  0.161 ( 2.787)  Data  0.001 ( 0.364)  Loss 1.1415e+00 (1.9788e+00)  Acc@1 100.00 ( 19.67)  Acc@5 100.00 ( 91.80)
Epoch: [20][ 70/900]  Time  0.160 ( 2.417)  Data  0.000 ( 0.314)  Loss 1.7265e+00 (2.0525e+00)  Acc@1   0.00 ( 18.31)  Acc@5 100.00 ( 88.73)
Epoch: [20][ 80/900]  Time  0.160 ( 2.138)  Data  0.000 ( 0.276)  Loss 2.1617e+00 (2.0397e+00)  Acc@1   0.00 ( 17.28)  Acc@5 100.00 ( 87.65)
Epoch: [20][ 90/900]  Time  0.158 ( 1.921)  Data  0.000 ( 0.246)  Loss 1.3998e+00 (1.9995e+00)  Acc@1   0.00 ( 17.58)  Acc@5 100.00 ( 89.01)
Epoch: [20][100/900]  Time  0.159 ( 1.746)  Data  0.001 ( 0.223)  Loss 1.5428e+00 (1.9539e+00)  Acc@1   0.00 ( 19.80)  Acc@5 100.00 ( 90.10)
Epoch: [20][110/900]  Time  0.157 ( 1.603)  Data  0.001 ( 0.203)  Loss 3.0696e+00 (2.0262e+00)  Acc@1   0.00 ( 18.92)  Acc@5   0.00 ( 88.29)
Epoch: [20][120/900]  Time  0.158 ( 1.484)  Data  0.000 ( 0.187)  Loss 2.6331e+00 (2.0204e+00)  Acc@1   0.00 ( 18.18)  Acc@5 100.00 ( 88.43)
Epoch: [20][130/900]  Time  0.161 ( 1.383)  Data  0.000 ( 0.174)  Loss 1.8139e+00 (2.0167e+00)  Acc@1   0.00 ( 19.08)  Acc@5 100.00 ( 87.79)
Epoch: [20][140/900]  Time  0.161 ( 1.296)  Data  0.001 ( 0.162)  Loss 2.1644e+00 (2.0445e+00)  Acc@1   0.00 ( 18.44)  Acc@5   0.00 ( 85.11)
Epoch: [20][150/900]  Time  0.158 ( 1.221)  Data  0.000 ( 0.152)  Loss 2.8062e+00 (2.0470e+00)  Acc@1   0.00 ( 17.88)  Acc@5   0.00 ( 84.11)
Epoch: [20][160/900]  Time  0.156 ( 1.155)  Data  0.001 ( 0.143)  Loss 1.7622e+00 (2.0375e+00)  Acc@1   0.00 ( 18.01)  Acc@5 100.00 ( 84.47)
Epoch: [20][170/900]  Time  0.159 ( 1.097)  Data  0.000 ( 0.135)  Loss 3.6050e-01 (2.0003e+00)  Acc@1 100.00 ( 20.47)  Acc@5 100.00 ( 84.80)
Epoch: [20][180/900]  Time  0.158 ( 1.045)  Data  0.000 ( 0.128)  Loss 1.6848e+00 (2.0566e+00)  Acc@1   0.00 ( 19.34)  Acc@5 100.00 ( 83.43)
Epoch: [20][190/900]  Time  0.160 ( 0.998)  Data  0.000 ( 0.121)  Loss 1.9842e+00 (2.0412e+00)  Acc@1   0.00 ( 19.90)  Acc@5 100.00 ( 83.77)
Epoch: [20][200/900]  Time  0.156 ( 0.957)  Data  0.000 ( 0.116)  Loss 1.4973e+00 (2.0434e+00)  Acc@1   0.00 ( 19.90)  Acc@5 100.00 ( 84.08)
Epoch: [20][210/900]  Time  0.158 ( 0.919)  Data  0.001 ( 0.111)  Loss 2.1533e+00 (2.0486e+00)  Acc@1   0.00 ( 19.43)  Acc@5 100.00 ( 84.36)
Epoch: [20][220/900]  Time  0.158 ( 0.884)  Data  0.000 ( 0.106)  Loss 1.7518e+00 (2.0428e+00)  Acc@1   0.00 ( 19.46)  Acc@5 100.00 ( 84.62)
Epoch: [20][230/900]  Time  0.158 ( 0.853)  Data  0.001 ( 0.102)  Loss 1.8696e+00 (2.0310e+00)  Acc@1   0.00 ( 20.78)  Acc@5 100.00 ( 84.42)
Epoch: [20][240/900]  Time  0.156 ( 0.824)  Data  0.001 ( 0.098)  Loss 1.0003e+00 (2.0360e+00)  Acc@1 100.00 ( 21.16)  Acc@5 100.00 ( 84.23)
Epoch: [20][250/900]  Time  0.159 ( 0.798)  Data  0.000 ( 0.094)  Loss 2.3924e+00 (2.0312e+00)  Acc@1   0.00 ( 21.12)  Acc@5   0.00 ( 83.67)
Epoch: [20][260/900]  Time  0.148 ( 0.773)  Data  0.000 ( 0.091)  Loss 1.6305e+00 (2.0256e+00)  Acc@1   0.00 ( 21.46)  Acc@5 100.00 ( 83.91)
Epoch: [20][270/900]  Time  0.156 ( 0.751)  Data  0.001 ( 0.088)  Loss 2.1856e+00 (2.0292e+00)  Acc@1   0.00 ( 21.03)  Acc@5   0.00 ( 82.66)
Epoch: [20][280/900]  Time  0.159 ( 0.730)  Data  0.001 ( 0.085)  Loss 2.7033e+00 (2.0430e+00)  Acc@1   0.00 ( 20.28)  Acc@5   0.00 ( 81.49)
Epoch: [20][290/900]  Time  0.159 ( 0.710)  Data  0.000 ( 0.082)  Loss 1.5022e+00 (2.0399e+00)  Acc@1   0.00 ( 20.27)  Acc@5 100.00 ( 81.10)
Epoch: [20][300/900]  Time  0.159 ( 0.692)  Data  0.001 ( 0.080)  Loss 2.3162e+00 (2.0365e+00)  Acc@1   0.00 ( 20.60)  Acc@5 100.00 ( 80.73)
Epoch: [20][310/900]  Time  0.160 ( 0.675)  Data  0.000 ( 0.078)  Loss 3.3808e+00 (2.0352e+00)  Acc@1   0.00 ( 20.90)  Acc@5   0.00 ( 80.71)
Epoch: [20][320/900]  Time  0.158 ( 0.659)  Data  0.001 ( 0.075)  Loss 1.6032e+00 (2.0371e+00)  Acc@1   0.00 ( 20.56)  Acc@5 100.00 ( 80.69)
Epoch: [20][330/900]  Time  0.157 ( 0.643)  Data  0.000 ( 0.073)  Loss 1.1670e+00 (2.0275e+00)  Acc@1   0.00 ( 20.24)  Acc@5 100.00 ( 81.27)
Epoch: [20][340/900]  Time  0.158 ( 0.629)  Data  0.001 ( 0.071)  Loss 6.7029e-01 (2.0336e+00)  Acc@1 100.00 ( 20.53)  Acc@5 100.00 ( 81.52)
Epoch: [20][350/900]  Time  0.159 ( 0.616)  Data  0.001 ( 0.070)  Loss 2.3698e+00 (2.0486e+00)  Acc@1   0.00 ( 19.94)  Acc@5 100.00 ( 80.63)
Epoch: [20][360/900]  Time  0.157 ( 0.603)  Data  0.000 ( 0.068)  Loss 2.1335e+00 (2.0412e+00)  Acc@1   0.00 ( 20.50)  Acc@5 100.00 ( 80.61)
Epoch: [20][370/900]  Time  0.160 ( 0.591)  Data  0.001 ( 0.066)  Loss 1.2511e+00 (2.0331e+00)  Acc@1 100.00 ( 21.02)  Acc@5 100.00 ( 80.86)
Epoch: [20][380/900]  Time  0.157 ( 0.580)  Data  0.000 ( 0.065)  Loss 3.0436e+00 (2.0418e+00)  Acc@1   0.00 ( 20.73)  Acc@5   0.00 ( 80.58)
Epoch: [20][390/900]  Time  0.159 ( 0.569)  Data  0.000 ( 0.063)  Loss 1.3161e+00 (2.0407e+00)  Acc@1 100.00 ( 20.72)  Acc@5 100.00 ( 81.07)
Epoch: [20][400/900]  Time  0.159 ( 0.559)  Data  0.000 ( 0.062)  Loss 2.5490e+00 (2.0370e+00)  Acc@1   0.00 ( 20.95)  Acc@5 100.00 ( 81.30)
Epoch: [20][410/900]  Time  0.157 ( 0.549)  Data  0.001 ( 0.061)  Loss 1.5901e+00 (2.0384e+00)  Acc@1   0.00 ( 20.92)  Acc@5 100.00 ( 81.27)
Epoch: [20][420/900]  Time  0.159 ( 0.540)  Data  0.000 ( 0.059)  Loss 1.7302e+00 (2.0333e+00)  Acc@1   0.00 ( 21.14)  Acc@5 100.00 ( 81.24)
Epoch: [20][430/900]  Time  0.158 ( 0.531)  Data  0.000 ( 0.058)  Loss 2.1498e+00 (2.0323e+00)  Acc@1   0.00 ( 20.88)  Acc@5 100.00 ( 81.21)
Epoch: [20][440/900]  Time  0.158 ( 0.523)  Data  0.001 ( 0.057)  Loss 1.7591e+00 (2.0276e+00)  Acc@1   0.00 ( 20.86)  Acc@5 100.00 ( 81.63)
Epoch: [20][450/900]  Time  0.157 ( 0.515)  Data  0.000 ( 0.056)  Loss 1.7298e+00 (2.0251e+00)  Acc@1   0.00 ( 20.84)  Acc@5 100.00 ( 82.04)
Epoch: [20][460/900]  Time  0.160 ( 0.507)  Data  0.001 ( 0.055)  Loss 2.7736e+00 (2.0277e+00)  Acc@1   0.00 ( 20.61)  Acc@5 100.00 ( 82.21)
Epoch: [20][470/900]  Time  0.162 ( 0.499)  Data  0.000 ( 0.054)  Loss 1.2765e+00 (2.0307e+00)  Acc@1 100.00 ( 20.59)  Acc@5 100.00 ( 81.95)
Epoch: [20][480/900]  Time  0.159 ( 0.492)  Data  0.000 ( 0.053)  Loss 2.0055e+00 (2.0271e+00)  Acc@1   0.00 ( 20.37)  Acc@5 100.00 ( 82.33)
Epoch: [20][490/900]  Time  0.159 ( 0.486)  Data  0.001 ( 0.052)  Loss 3.1931e+00 (2.0320e+00)  Acc@1   0.00 ( 20.37)  Acc@5   0.00 ( 82.08)
Epoch: [20][500/900]  Time  0.160 ( 0.479)  Data  0.000 ( 0.051)  Loss 2.2306e+00 (2.0368e+00)  Acc@1   0.00 ( 20.36)  Acc@5 100.00 ( 81.84)
Epoch: [20][510/900]  Time  0.157 ( 0.473)  Data  0.001 ( 0.050)  Loss 2.0753e+00 (2.0380e+00)  Acc@1   0.00 ( 20.16)  Acc@5 100.00 ( 82.00)
Epoch: [20][520/900]  Time  0.158 ( 0.467)  Data  0.000 ( 0.049)  Loss 1.6929e+00 (2.0398e+00)  Acc@1   0.00 ( 19.96)  Acc@5 100.00 ( 82.15)
Epoch: [20][530/900]  Time  0.157 ( 0.461)  Data  0.001 ( 0.049)  Loss 1.6216e+00 (2.0457e+00)  Acc@1   0.00 ( 19.59)  Acc@5 100.00 ( 82.11)
Epoch: [20][540/900]  Time  0.159 ( 0.455)  Data  0.000 ( 0.048)  Loss 2.0313e+00 (2.0450e+00)  Acc@1   0.00 ( 19.41)  Acc@5 100.00 ( 81.89)
Epoch: [20][550/900]  Time  0.158 ( 0.450)  Data  0.000 ( 0.047)  Loss 3.1055e+00 (2.0460e+00)  Acc@1   0.00 ( 19.42)  Acc@5   0.00 ( 81.67)
Epoch: [20][560/900]  Time  0.159 ( 0.445)  Data  0.000 ( 0.046)  Loss 2.0760e+00 (2.0461e+00)  Acc@1   0.00 ( 19.25)  Acc@5 100.00 ( 81.64)
Epoch: [20][570/900]  Time  0.159 ( 0.440)  Data  0.000 ( 0.046)  Loss 1.9432e+00 (2.0523e+00)  Acc@1   0.00 ( 18.91)  Acc@5 100.00 ( 81.61)
Epoch: [20][580/900]  Time  0.158 ( 0.435)  Data  0.001 ( 0.045)  Loss 1.7295e+00 (2.0495e+00)  Acc@1   0.00 ( 19.10)  Acc@5 100.00 ( 81.41)
Epoch: [20][590/900]  Time  0.155 ( 0.430)  Data  0.000 ( 0.045)  Loss 1.5650e+00 (2.0458e+00)  Acc@1   0.00 ( 19.29)  Acc@5 100.00 ( 81.39)
Epoch: [20][600/900]  Time  0.159 ( 0.426)  Data  0.000 ( 0.044)  Loss 2.7813e+00 (2.0452e+00)  Acc@1   0.00 ( 19.30)  Acc@5 100.00 ( 81.70)
Epoch: [20][610/900]  Time  0.158 ( 0.421)  Data  0.000 ( 0.043)  Loss 1.2349e+00 (2.0371e+00)  Acc@1 100.00 ( 19.31)  Acc@5 100.00 ( 82.00)
Epoch: [20][620/900]  Time  0.158 ( 0.417)  Data  0.000 ( 0.043)  Loss 1.7804e+00 (2.0417e+00)  Acc@1   0.00 ( 19.48)  Acc@5 100.00 ( 81.96)
Epoch: [20][630/900]  Time  0.158 ( 0.413)  Data  0.001 ( 0.042)  Loss 1.2240e+00 (2.0431e+00)  Acc@1 100.00 ( 19.33)  Acc@5 100.00 ( 81.93)
Epoch: [20][640/900]  Time  0.157 ( 0.409)  Data  0.000 ( 0.042)  Loss 2.4593e+00 (2.0373e+00)  Acc@1   0.00 ( 19.81)  Acc@5 100.00 ( 81.90)
Epoch: [20][650/900]  Time  0.157 ( 0.405)  Data  0.000 ( 0.041)  Loss 1.9445e+00 (2.0415e+00)  Acc@1   0.00 ( 19.82)  Acc@5 100.00 ( 81.87)
Epoch: [20][660/900]  Time  0.159 ( 0.402)  Data  0.000 ( 0.041)  Loss 1.4294e+00 (2.0369e+00)  Acc@1   0.00 ( 19.97)  Acc@5 100.00 ( 82.00)
Epoch: [20][670/900]  Time  0.159 ( 0.398)  Data  0.001 ( 0.040)  Loss 9.3780e-01 (2.0358e+00)  Acc@1 100.00 ( 19.97)  Acc@5 100.00 ( 82.12)
Epoch: [20][680/900]  Time  0.160 ( 0.394)  Data  0.001 ( 0.040)  Loss 1.7501e+00 (2.0380e+00)  Acc@1   0.00 ( 19.82)  Acc@5 100.00 ( 81.94)
Epoch: [20][690/900]  Time  0.157 ( 0.391)  Data  0.000 ( 0.039)  Loss 2.0664e+00 (2.0381e+00)  Acc@1   0.00 ( 19.54)  Acc@5 100.00 ( 81.77)
Epoch: [20][700/900]  Time  0.156 ( 0.388)  Data  0.000 ( 0.039)  Loss 2.0604e+00 (2.0368e+00)  Acc@1   0.00 ( 19.69)  Acc@5 100.00 ( 81.88)
Epoch: [20][710/900]  Time  0.156 ( 0.385)  Data  0.001 ( 0.038)  Loss 2.5807e+00 (2.0388e+00)  Acc@1   0.00 ( 19.69)  Acc@5 100.00 ( 81.86)
Epoch: [20][720/900]  Time  0.159 ( 0.381)  Data  0.001 ( 0.038)  Loss 2.1288e+00 (2.0403e+00)  Acc@1   0.00 ( 19.69)  Acc@5 100.00 ( 82.11)
Epoch: [20][730/900]  Time  0.161 ( 0.378)  Data  0.000 ( 0.037)  Loss 1.3987e+00 (2.0423e+00)  Acc@1   0.00 ( 19.56)  Acc@5 100.00 ( 82.08)
Epoch: [20][740/900]  Time  0.158 ( 0.375)  Data  0.000 ( 0.037)  Loss 1.5643e+00 (2.0443e+00)  Acc@1   0.00 ( 19.43)  Acc@5 100.00 ( 81.92)
Epoch: [20][750/900]  Time  0.157 ( 0.373)  Data  0.000 ( 0.037)  Loss 2.3474e+00 (2.0450e+00)  Acc@1   0.00 ( 19.31)  Acc@5   0.00 ( 81.62)
Epoch: [20][760/900]  Time  0.156 ( 0.370)  Data  0.000 ( 0.036)  Loss 1.7025e+00 (2.0420e+00)  Acc@1   0.00 ( 19.19)  Acc@5 100.00 ( 81.73)
Epoch: [20][770/900]  Time  0.159 ( 0.367)  Data  0.001 ( 0.036)  Loss 2.3098e+00 (2.0422e+00)  Acc@1   0.00 ( 19.20)  Acc@5 100.00 ( 81.58)
Epoch: [20][780/900]  Time  0.156 ( 0.364)  Data  0.000 ( 0.036)  Loss 1.9435e+00 (2.0425e+00)  Acc@1   0.00 ( 19.08)  Acc@5 100.00 ( 81.18)
Epoch: [20][790/900]  Time  0.160 ( 0.362)  Data  0.000 ( 0.035)  Loss 1.6258e+00 (2.0400e+00)  Acc@1   0.00 ( 19.09)  Acc@5 100.00 ( 81.29)
Epoch: [20][800/900]  Time  0.157 ( 0.359)  Data  0.001 ( 0.035)  Loss 1.1790e+00 (2.0414e+00)  Acc@1 100.00 ( 18.98)  Acc@5 100.00 ( 81.27)
Epoch: [20][810/900]  Time  0.159 ( 0.357)  Data  0.001 ( 0.035)  Loss 2.2912e+00 (2.0404e+00)  Acc@1   0.00 ( 18.99)  Acc@5   0.00 ( 81.26)
Epoch: [20][820/900]  Time  0.160 ( 0.354)  Data  0.001 ( 0.034)  Loss 2.4542e+00 (2.0406e+00)  Acc@1   0.00 ( 19.00)  Acc@5   0.00 ( 81.00)
Epoch: [20][830/900]  Time  0.158 ( 0.352)  Data  0.000 ( 0.034)  Loss 1.2103e+00 (2.0384e+00)  Acc@1 100.00 ( 19.01)  Acc@5 100.00 ( 80.99)
Epoch: [20][840/900]  Time  0.158 ( 0.350)  Data  0.001 ( 0.034)  Loss 1.6468e+00 (2.0388e+00)  Acc@1   0.00 ( 18.79)  Acc@5 100.00 ( 80.98)
Epoch: [20][850/900]  Time  0.156 ( 0.347)  Data  0.001 ( 0.033)  Loss 1.7870e+00 (2.0385e+00)  Acc@1   0.00 ( 18.92)  Acc@5 100.00 ( 80.85)
Epoch: [20][860/900]  Time  0.159 ( 0.345)  Data  0.000 ( 0.033)  Loss 2.2311e+00 (2.0392e+00)  Acc@1   0.00 ( 18.93)  Acc@5 100.00 ( 80.72)
Epoch: [20][870/900]  Time  0.158 ( 0.343)  Data  0.001 ( 0.033)  Loss 1.6941e+00 (2.0374e+00)  Acc@1   0.00 ( 18.94)  Acc@5 100.00 ( 80.83)
Epoch: [20][880/900]  Time  0.159 ( 0.341)  Data  0.001 ( 0.032)  Loss 2.3661e+00 (2.0384e+00)  Acc@1   0.00 ( 18.84)  Acc@5 100.00 ( 80.82)
Epoch: [20][890/900]  Time  0.159 ( 0.339)  Data  0.001 ( 0.032)  Loss 2.3517e+00 (2.0357e+00)  Acc@1   0.00 ( 18.86)  Acc@5 100.00 ( 81.03)
Epoch: [20] completed, elapsed time 303.842 seconds
Test: [ 0/90]  Time  2.105 ( 2.105)  Loss 1.7548e+00 (1.7548e+00)  Acc@1   0.00 (  0.00)  Acc@5 100.00 (100.00)
Test: [10/90]  Time  0.084 ( 0.259)  Loss 1.5891e+00 (1.6303e+00)  Acc@1   0.00 (  0.00)  Acc@5 100.00 (100.00)
Test: [20/90]  Time  0.081 ( 0.176)  Loss 1.7160e+00 (1.6740e+00)  Acc@1   0.00 (  0.00)  Acc@5 100.00 (100.00)
Test: [30/90]  Time  0.109 ( 0.149)  Loss 1.7012e+00 (1.6648e+00)  Acc@1   0.00 (  3.23)  Acc@5 100.00 (100.00)
Test: [40/90]  Time  0.083 ( 0.147)  Loss 1.7273e+00 (1.6799e+00)  Acc@1   0.00 (  2.44)  Acc@5 100.00 (100.00)
Test: [50/90]  Time  0.086 ( 0.135)  Loss 2.1467e+00 (1.7406e+00)  Acc@1   0.00 (  1.96)  Acc@5 100.00 (100.00)
Test: [60/90]  Time  0.091 ( 0.127)  Loss 1.4457e+00 (1.7972e+00)  Acc@1 100.00 (  3.28)  Acc@5 100.00 (100.00)
Test: [70/90]  Time  0.086 ( 0.121)  Loss 1.4470e+00 (1.7477e+00)  Acc@1 100.00 ( 16.90)  Acc@5 100.00 (100.00)
Test: [80/90]  Time  0.082 ( 0.117)  Loss 2.5285e+00 (1.7892e+00)  Acc@1   0.00 ( 19.75)  Acc@5   0.00 ( 92.59)
 * Acc@1 17.778 Acc@5 84.444
saved checkpoint to:  models/Tnew6c20e/checkpoint.pth.tar

Accuracy does not increase from epoch 7 until 200. Even after 200 Epochs, my overall accuracy is 23.33%
Now please guide me in this regard. My highest overall accuracy is 27%.

waseemsofficial · July 20, 2022, 11:48am

can u please guide me on how to make changes in models or layers in the pre-trained model? or any other suggestion on how to increase accuracy.

waseemsofficial · July 22, 2022, 9:47am

Please guide me about the layers of the Model. or how we can change parameters and layers
Thanks

dusty_nv · July 22, 2022, 1:25pm

Hi @waseemsofficial, you can change the network architecture with the --arch argument when you run train.py. Here are the different resnet variants in torchvision that you can try:

resnet18
resnet34
resnet50
resnet101
resnet152

For example, you can use --arch=resnet34 or --arch=resnet50 to add more layers to resnet. However, since your training accuracy is so low to begin with, my guess is that your dataset may need improved as well. It sounds like you only have 15 images per class, whereas you should have a lot more than that.

system · August 10, 2022, 6:29am

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.

Topic		Replies	Views
Jetson AI Fundamentals - S3E3 - Training Image Classification Models Jetson Nano jetson-inference	6	633	February 21, 2023
Re-training ResNet-18 model on Jetson nano \| how many epochs? The model shows only one class on everything after re-training Jetson Nano jetson-inference , jetson-nano	7	1806	May 12, 2022
Data corruption when running train_ssd script Jetson Nano python , training	10	912	September 12, 2022
Training Custom FasterRCNN resnet50 Object detection issue TAO Toolkit	9	1118	October 12, 2021
How to plot graph after training a model? Jetson Nano ai-training	12	1397	June 30, 2022
Low confidence scores of cat and dog in jetson-inference classification Jetson Nano jetson-inference , ai-training	4	899	October 18, 2021
Train custom object detectio model Jetson Nano ai-training	12	3027	October 18, 2021
Faster RCNN ResNet-101 Problems TAO Toolkit	20	1104	October 12, 2021
Training with "train_ssd.py" - error at the end of the dataset Jetson AGX Xavier	6	1246	October 18, 2021
Issue in experimenting with transfer learning Jetson Nano ai-training	10	688	October 18, 2021

Testing accuracy is too low on ResNET 18 pre tarined model

Related topics