Models give same age and gender for different images

[jvdrag]

Using the pre-trained model available in this repository got me to this:

Loading dataset: HuggingFaceM4/FairFace config: 0.25 /usr/local/envs/myenv/lib/python3.7/site-packages/datasets/load.py:1763: FutureWarning: 'ignore_verifications' was deprecated in favor of 'verification_mode' in version 2.9.1 and will be removed in 3.0.0. You can remove this warning by passing 'verification_mode=no_checks' instead. FutureWarning, Found cached dataset parquet (/root/.cache/huggingface/datasets/HuggingFaceM4___parquet/0.25-1a13370f29aa8c70/0.0.0/14a00e99c0d15a23649d0db8944380ac81082d4b021f398733dd84f3a6c569a7) 100% 2/2 [00:00<00:00, 218.18it/s] Dataset loaded successfully. Dataset({ features: ['image', 'age', 'gender'], num_rows: 173488 }) Building model (using build_net3 from C3AE_expand.py architecture)...

Model: "C3AE_Expand_Net"

Weights loaded successfully. !!!!! Found NaN in weights, replacing with 0. Layer sample (first few elements): [-0.4407964 -0.01276569 0.31867692 -0.36723265 0.09467949] !!!!! Found NaN in weights, replacing with 0. Layer sample (first few elements): [0.95068 2.0552835 1.1609683 0.68024117 1.2869499 ] !!!!! Found NaN in weights, replacing with 0. Layer sample (first few elements): [-0.9160388 -1.0657301 -1.5303322 -0.6400793 -0.7105152] !!!!! Found NaN in weights, replacing with 0. Layer sample (first few elements): [-217.23065 -214.79196 -6.020306 487.3062 -47.080265] !!!!! Found NaN in weights, replacing with 0. Layer sample (first few elements): [46061.426 46790.33 1892.5228 98859.266 22331.162 ] --- Weights refreshed (NaNs replaced with 0) ---

--- Creating Intermediate Models for Debugging --- Intermediate models created.

Processing 100 examples... Processing example 1/100

--- Intermediate Activation Check (Example 0) --- Base Model Output (v1) - Shape: (1, 512), Mean: -1.8121, Std: 6.1161 Normalized cfeat Output - Shape: (1, 512), Mean: -0.0030, Std: 0.0441

--- Intermediate Activation Check (Example 1) --- Base Model Output (v1) - Shape: (1, 512), Mean: -1.8121, Std: 6.1163 Normalized cfeat Output - Shape: (1, 512), Mean: -0.0030, Std: 0.0441

--- Intermediate Activation Check (Example 2) --- Base Model Output (v1) - Shape: (1, 512), Mean: -1.8121, Std: 6.1164 Normalized cfeat Output - Shape: (1, 512), Mean: -0.0030, Std: 0.0441

Processing example 51/100 Finished processing. 100 examples successfully processed.

--- Prediction Results Sample (First 15) --- original_age_label original_age_range original_gender predicted_age
0 6 50-59 Male -0.700401
1 4 30-39 Female -0.700441
2 1 3-9 Female -0.700463
3 3 20-29 Female -0.700420
4 3 20-29 Female -0.700464
5 3 20-29 Male -0.700429
6 5 40-49 Male -0.700477
7 4 30-39 Female -0.700402
8 2 10-19 Male -0.700436
9 4 30-39 Male -0.700414
10 6 50-59 Male -0.700444
11 3 20-29 Male -0.700460
12 3 20-29 Male -0.700452
13 2 10-19 Male -0.700386
14 7 60-69 Female -0.700453

predicted_gender
0 Male
1 Male
2 Male
3 Male
4 Male
5 Male
6 Male
7 Male
8 Male
9 Male
10 Male
11 Male
12 Male
13 Male
14 Male

Age prediction correct within range: 0/100 (0.00%) Gender prediction correct: 48/100 (48.00%)